#### Beyond the hype: Capturing the potential of AI and gen AI in tech, media, and telecom

February 2024


-----

**Beyond the hype:**
**Capturing the potential**
**of AI and gen AI in tech,**
**media, and telecom**

February 2024


-----

##### Contents

Introduction: The promise and the challenge of generative AI 2

4
**State of the Art**

The economic potential of generative AI 5

Making the most of the generative AI opportunity: Six questions for CEOs 33

38
**Sector View: Telecom Operators**

The AI-native telco: Radical transformation to thrive in turbulent times 39

How generative AI could revitalize profitability for telcos 48

Generative AI use cases: A guide to developing the telco of the future 60

Tech talent in transition: Seven technology trends reshaping telcos 70

81
**Deploying Gen AI**

The organization of the future: Enabled by gen AI, driven by people 82

The data dividend: Fueling generative AI 91

Technology’s generational moment with generative AI: A CIO and CTO guide 101

As gen AI advances, regulators—and risk functions—rush to keep pace 113

119
**What the Future Holds**

Six major gen AI trends that will shape 2024’s agenda 120

Appendix: Generative AI solutions in action 125

Glossary 127


-----

**Introduction: The promise and**
**the challenge of generative AI**

The emergence of generative AI (gen AI) presents both a challenge and a significant opportunity for leaders looking
to steer their organizations into the future. How big is the opportunity? McKinsey research estimates that gen AI
could add to the economy between $2.6 trillion and $4.4 trillion annually while increasing the impact of all artificial
intelligence by 15 to 40 percent. In the technology, media, and telecommunications (TMT) space, new gen AI use
cases are expected to unleash between $380 billion and $690 billion in impact—$60 billion to $100 billion in
telecommunications, $80 billion to $130 billion in media, and about $240 billion to $460 billion in high tech. In
fact, it seems possible that within the next three years, anything not connected to AI will be considered obsolete or
ineffective.

Some leaders are moving to seize the moment and implement gen AI in their organizations at scale, but others remain
in the pilot stage, and some have yet to decide what to do. If companies are to remain competitive and relevant in the
coming years, it is essential that executives understand the potential impact of gen AI and develop the strategies
necessary to incorporate it into their operations. Such strategies would involve an AI-native transformation, focused
on building and managing the adoption of gen AI. McKinsey has conducted extensive research into how to embed
gen AI to ensure that the technology delivers meaningful value. We’ve also spent much of the past year working with
clients to create and then implement gen AI road maps. That combination of research and hands-on experience has
allowed us to identify more than 100 gen AI use cases in TMT across seven business domains.[1]

Our experience working with clients already indicates the potential for telcos to achieve significant impact with
gen AI across all key functions. The largest share of total impact will likely be in customer care and sales, which
together would account for approximately 70 percent of total impact; network operations, IT, and support functions
would round out the rest. The technology already is showing meaningful impact in enhancing interactions between
employees and customers: the personalization of products and campaigns, improvements in sales effectiveness, and
a reduction in time to market can spark a potential revenue increase of 3 to 5 percent. Customer care interactions—
where as much as 50 percent of activity could be automated—have potential for a 30 to 45 percent increase in
productivity while improving the customer experience and customer satisfaction scores. On the labor side, up to 70
percent of repetitive work activities could be automated via gen AI to improve productivity. There is also potential for
new efficiencies in knowledge search, validation, and synthesis, where some 60 percent of activity has the potential
for automation. And gen AI tools could boost developer productivity by 20 to 45 percent.

These areas provide rich soil for use cases. More challenging will be to go from sketching a road map to building
proofs of concept to scaling successfully and capturing impact. Years of experience in designing and implementing
digital transformations have taught us a lot, but gen AI’s nature and speed of disruption are creating a new layer of
uncertainty.

Becoming an AI-native organization at scale involves making the most of technology, data, and governance. Success
follows when leaders embrace an operating model that leverages the strengths of both humans and machines; is
rooted in agility, flexibility, and continuous learning; and is supported by strong data and analytics talent. Another
condition of success is to invest in data quality and quantity, focusing on the data life cycle to ensure high-quality
information for training the gen AI model. Building capabilities into the data architecture, such as vector databases
and data pre- and post-processing pipelines, will enable the development of use cases. Talent, data, technology,
governance—none of these can be an afterthought.

¹ Marketing and digital, sales and channels, customer care, customer strategy, support, additional areas, and new businesses.


-----

Successful implementations share a clear vision and decisive approach. We advise that financial plans maintain or
increase gen AI budgets over the next year. These budgets should include resources dedicated to gen AI for the shaping
and crafting of bespoke solutions (for example, training large language models with telco-specific data, rather than
implementing off-the-shelf ones) or partnerships with IT vendors to accelerate the timeline for implementation.

The AI journey has been shown to contain many challenges and learning opportunities, such as preparing and shifting
an organization’s culture, finding data sets of significant size, and addressing the interpretability of the outputs provided
by models. Leaders should expect such daunting challenges as a shortage of talent, lack of organizational commitment
and prioritization (including among C-level executives), and difficulties in justifying ROI for certain business cases, all
amid a changing regulatory and ethics landscape that creates further uncertainty. But daunting does not have to mean
impossible. Developing a system of protocols and guardrails (such as building “moderation” models to check outputs
for different risks and ensure users receive consistent responses) will be a crucial step toward mitigating the new risks
introduced by gen AI. Another key will be change management—involving end users in the model development process
and deeply embedding technology into their operations.

This collection presents McKinsey’s top insights on gen AI, providing a detailed examination of this technology’s
transformative potential for organizations. It offers top management guidance on how to prepare for the implementation
of gen AI and explores the implications of gen AI’s use by the TMT industries, especially telecommunications. The
collection covers the essential requirements for deploying gen AI, including organizational readiness, data management,
and technological considerations. It also emphasizes the importance of effectively managing risks associated with gen
AI implementation. Furthermore, this compilation offers an overview of the future developments and advancements
expected in the field of generative AI.

Gen AI will continue to evolve. New capabilities, such as the ability to analyze and comprehend images or audio, and an
expanding ecosystem with marketplaces for GPT (generative pretrained transformers), are constantly emerging. For
leaders, the stakes are high. But so are the opportunities. The next move from TMT players will define how they move
from isolated cases to implementations at scale, from hype to impact.


**Alex Singla**
Senior Partner
Managing Partner
QuantumBlack
AI by McKinsey


**Alexander Sukharevsky**
Senior Partner
Managing Partner
QuantumBlack
AI by McKinsey


**Brendan Gaffey**
Senior Partner
Global Leader
TMT Practice


**Noshir Kaka**
Senior Partner
Global Leader
TMT Practice


**Peter Dahlström**
Senior Partner
Europe Leader
TMT Practice

**Tomás Lajous**
Senior Partner
AI and Gen AI Leader
TMT Practice


**Andrea Travasoni**
Senior Partner
Global Leader
Telecom Operators
TMT Practice

**Benjamim Vieira**
Senior Partner
Digital and Analytics Leader
TMT Practice


**Venkat Atluri**
Senior Partner
Global Leader
Telecom Operators
TMT Practice

**Víctor García de la Torre**
Associate Partner
TMT Practice


-----

### State of the art


-----

### The economic potential of generative AI

The next productivity frontier

June 2023

Authors

Michael Chui

Eric Hazan

Roger Roberts

Alex Singla

Kate Smaje

Alexander Sukharevsky

Lareina Yee

Rodney Zemmel


-----

## 1


**Generative AI as a**
**technology catalyst**

To grasp what lies ahead requires an understanding of the breakthroughs that have enabled the rise
of generative AI, which were decades in the making. ChatGPT, GitHub Copilot, Stable Diff usion, and
other generative AI tools that have captured current public attention are the result of signifi cant levels
of investment in recent years that have helped advance machine learning and deep learning. This
investment undergirds the AI applications embedded in many of the products and services we use
every day.

But because AI has permeated our lives incrementally—through everything from the tech powering
our smartphones to autonomous-driving features on cars to the tools retailers use to surprise and
delight consumers—its progress was almost imperceptible. Clear milestones, such as when AlphaGo,
an AI-based program developed by DeepMind, defeated a world champion Go player in 2016, were
celebrated but then quickly faded from the public’s consciousness.

ChatGPT and its competitors have captured the imagination of people around the world in a way
AlphaGo did not, thanks to their broad utility—almost anyone can use them to communicate and
create—and preternatural ability to have a conversation with a user. The latest generative AI
applications can perform a range of routine tasks, such as the reorganization and classifi cation

_This article is excerpted from the full McKinsey report, The economic potential of generative AI: The_
next productivity frontier. To read the full report, including details about the research, appendix, and
_acknowledgements, visit mck.co/genai._


-----

of data. But it is their ability to write text, compose music, and create digital art that has garnered
headlines and persuaded consumers and households to experiment on their own. As a result, a
broader set of stakeholders are grappling with generative AI’s impact on business and society but
without much context to help them make sense of it.

**How did we get here? Gradually, then all of a sudden**

For the purposes of this report, we define generative AI as applications typically built using foundation
models. These models contain expansive artificial neural networks inspired by the billions of neurons
connected in the human brain. Foundation models are part of what is called deep learning, a term
that alludes to the many deep layers within neural networks. Deep learning has powered many of
the recent advances in AI, but the foundation models powering generative AI applications are a step
change evolution within deep learning. Unlike previous deep learning models, they can process
extremely large and varied sets of unstructured data and perform more than one task.

Foundation models have enabled new capabilities and vastly improved existing ones across a broad
range of modalities, including images, video, audio, and computer code. AI trained on these models
can perform several functions; it can classify, edit, summarize, answer questions, and draft new
content, among other tasks.

Continued innovation will also bring new challenges. For example, the computational power required
to train generative AI with hundreds of billions of parameters threatens to become a bottleneck in
development.¹ Further, there’s a significant move—spearheaded by the open-source community and
spreading to the leaders of generative AI companies themselves—to make AI more responsible, which
could increase its costs.

Nonetheless, funding for generative AI, though still a fraction of total investments in artificial
intelligence, is significant and growing rapidly—reaching a total of $12 billion in the first five months
of 2023 alone. Venture capital and other private external investments in generative AI increased by
an average compound growth rate of 74 percent annually from 2017 to 2022. During the same period,
investments in artificial intelligence overall rose annually by 29 percent, albeit from a higher base.

The rush to throw money at all things generative AI reflects how quickly its capabilities have
developed. ChatGPT was released in November 2022. Four months later, OpenAI released a new
large language model, or LLM, called GPT-4 with markedly improved capabilities.² Similarly, by May
2023, Anthropic’s generative AI, Claude, was able to process 100,000 tokens of text, equal to about
75,000 words in a minute—the length of the average novel—compared with roughly 9,000 tokens
when it was introduced in March 2023.³ And in May 2023, Google announced several new features
powered by generative AI, including Search Generative Experience and a new LLM called PaLM 2 that
will power its Bard chatbot, among other Google products.⁴

From a geographic perspective, external private investment in generative AI, mostly from tech
giants and venture capital firms, is largely concentrated in North America, reflecting the continent’s
current domination of the overall AI investment landscape. Generative AI–related companies based
in the United States raised about $8 billion from 2020 to 2022, accounting for 75 percent of total
investments in such companies during that period.⁵

Generative AI has stunned and excited the world with its potential for reshaping how knowledge work
gets done in industries and business functions across the entire economy. Across functions such
as sales and marketing, customer operations, and software development, it is poised to transform
roles and boost performance. In the process, it could unlock trillions of dollars in value across sectors
from banking to life sciences. We have used two overlapping lenses in this report to understand


-----

## 2


**Generative AI use**
**cases across functions**
**and industries**

the potential for generative AI to create value for companies and alter the workforce. The
following sections share our initial findings.


-----

Exhibit 1

The potential impact of generative AI can be evaluated through two lenses.


Lens 1


Lens 2


_Total economic_ _Labor productivity potential_
_potential of 60-plus_ _across ~2,100 detailed work_
_organizational use_ _activities performed by_
_cases1_ _global workforce_

Cost impacts
of use cases

Revenue
impacts of
use cases1


1For quantitative analysis, revenue impacts were recast as productivity increases on the corresponding spend in order to maintain comparability with cost
impacts and not to assume additional growth in any particular market.

McKinsey & Company

Generative AI is a step change in the evolution of artifi cial intelligence. As companies
rush to adapt and implement it, understanding the technology’s potential to deliver value
to the economy and society at large will help shape critical decisions. We have used two
complementary lenses to determine where generative AI with its current capabilities could
deliver the biggest value and how big that value could be (Exhibit 1).

The fi rst lens scans use cases for generative AI that organizations could adopt. We defi ne
a “use case” as a targeted application of generative AI to a specifi c business challenge,
resulting in one or more measurable outcomes. For example, a use case in marketing is the
application of generative AI to generate creative content such as personalized emails, the
measurable outcomes of which potentially include reductions in the cost of generating such
content and increases in revenue from the enhanced eff ectiveness of higher-quality content
at scale. We identifi ed 63 generative AI use cases spanning 16 business functions that could
deliver total value in the range of $2.6 trillion to $4.4 trillion in economic benefi ts annually
when applied across industries.

That would add 15 to 40 percent to the $11.0 trillion to $17.7 trillion of economic value that we
now estimate nongenerative artifi cial intelligence and analytics could unlock. (Our previous
estimate from 2017 was that AI could deliver $9.5 trillion to $15.4 trillion in economic value.)

Our second lens complements the fi rst by analyzing generative AI’s potential impact on
the work activities required in some 850 occupations. We modeled scenarios to estimate
when generative AI could perform each of more than 2,100 “detailed work activities”—


-----

such as “communicating with others about operational plans or activities”—that make up
those occupations across the world economy. This enables us to estimate how the current
capabilities of generative AI could aff ect labor productivity across all work currently done by
the global workforce.

Some of this impact will overlap with cost reductions in the use case analysis described
above, which we assume are the result of improved labor productivity. Netting out this

Exhibit 2

Generative AI could create additional value potential above what
could be unlocked by other AI and analytics.

AI’s potential impact on the global economy, $ trillion

17.1–25.6


13.6–22.1

6.1–7.9
2.6–4.4

11.0–17.7

**~15–40%** **~35–70%**
incremental incremental
economic impact economic impact


Advanced analytics,
traditional machine
learning, and deep
learning[1]


New generative Total use
AI use cases case-driven
potential


All worker productivity
enabled by generative
AI, including in use
cases


Total AI
economic
potential


1Updated use case estimates from "Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.

McKinsey & Company


-----

overlap, the total economic benefits of generative AI—including the major use cases we
explored and the myriad increases in productivity that are likely to materialize when the
technology is applied across knowledge workers’ activities—amounts to $6.1 trillion to
$7.9 trillion annually (Exhibit 2).

While generative AI is an exciting and rapidly advancing technology, the other applications of
AI discussed in our previous report continue to account for the majority of the overall potential
value of AI. Traditional advanced-analytics and machine learning algorithms are highly

Box 1

**How we estimated the value potential of generative AI use cases**

To assess the potential value of generative AI, a customer service use case but not in a use
we updated a proprietary McKinsey database of case optimizing a logistics network, where value
potential AI use cases and drew on the experience primarily arises from quantitative analysis.
of more than 100 experts in industries and their

We then estimated the potential annual value

business functions.[1] Our updates examined

of these generative AI use cases if they were

use cases of generative AI—specifically, how

adopted across the entire economy. For use

generative AI techniques (primarily transformer
cases aimed at increasing revenue, such as some

based neural networks) can be used to solve

of those in sales and marketing, we estimated

problems not well addressed by previous

the economy-wide value generative AI could

technologies.

deliver by increasing the productivity of sales and

We analyzed only use cases for which generative marketing expenditures.
AI could deliver a significant improvement in the

Our estimates are based on the structure of the

outputs that drive key value. In particular, our

global economy in 2022 and do not consider the

estimates of the primary value the technology

value generative AI could create if it produced

could unlock do not include use cases for which

entirely new product or service categories.

the sole benefit would be its ability to use natural
language. For example, natural-language
capabilities would be the key driver of value in

1 “Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.


effective at performing numerical and optimization tasks such as predictive modeling, and
they continue to find new applications in a wide range of industries. However, as generative AI
continues to develop and mature, it has the potential to open wholly new frontiers in creativity
and innovation. It has already expanded the possibilities of what AI overall can achieve (see
Box 1, “How we estimated the value potential of generative AI use cases”).

In this chapter, we highlight the value potential of generative AI across two dimensions:
business function and modality.


-----

**Value potential by function**

While generative AI could have an impact on most business functions, a few stand out when
measured by the technology’s impact as a share of functional cost (Exhibit 3). Our analysis
of 16 business functions identifi ed just four—customer operations, marketing and sales,
software engineering, and research and development—that could account for approximately
75 percent of the total annual value from generative AI use cases.

Web <2023>

Exhibit 3<Vivatech full report>

Exhibit <3> of <16>
Using generative AI in just a few functions could drive most of the technology’s
impact across potential corporate use cases.

500

400

300

Impact, $ billion

200

100

0

|Col1|Represent ~75% of total annual impact of generative AI|Col3|Col4|Col5|
|---|---|---|---|---|
||Sales Software engineering Marketing (for corporate IT) Software engineering (for product development) Customer operations Product R&D1 Supply chain Manufacturing Finance Risk and compliance Talent and organization (incl HR) Procurement management Corporate IT1 Legal Strategy Pricing||||
||||||
||||||
||||||
||||||
||||||


0 10 20 30 40

Impact as a percentage of functional spend, %

Note: Impact is averaged.
¹Excluding software engineering.
Source: Comparative Industry Service (CIS), IHS Markit; Oxford Economics; McKinsey Corporate and Business Functions database; McKinsey Manufacturing
and Supply Chain 360; McKinsey Sales Navigator; Ignite, a McKinsey database; McKinsey analysis

McKinsey & Company

Notably, the potential value of using generative AI for several functions that were prominent
in our previous sizing of AI use cases, including manufacturing and supply chain functions,
is now much lower.⁶ This is largely explained by the nature of generative AI use cases, which
exclude most of the numerical and optimization applications that were the main value drivers
for previous applications of AI.


-----

Generative AI as a virtual expert
In addition to the potential value generative AI can deliver in function-specific use cases,
the technology could drive value across an entire organization by revolutionizing internal
knowledge management systems. Generative AI’s impressive command of natural-language
processing can help employees retrieve stored internal knowledge by formulating queries
in the same way they might ask a human a question and engage in continuing dialogue. This
could empower teams to quickly access relevant information, enabling them to rapidly make
better-informed decisions and develop effective strategies.

In 2012, the McKinsey Global Institute (MGI) estimated that knowledge workers spent about
a fifth of their time, or one day each workweek, searching for and gathering information. If
generative AI could take on such tasks, increasing the efficiency and effectiveness of the
workers doing them, the benefits would be huge. Such virtual expertise could rapidly “read”
vast libraries of corporate information stored in natural language and quickly scan source
material in dialogue with a human who helps fine-tune and tailor its research, a more scalable
solution than hiring a team of human experts for the task.

Following are examples of how generative AI could produce operational benefits as a virtual
expert in a handful of use cases.

**In addition to the potential**
**value generative AI can**
**deliver in function-specific**
**use cases, the technology**
**could drive value across**
**an entire organization**
**by revolutionizing**
**internal knowledge**
**management systems.**


-----

_Customer operations_
Generative AI has the potential to revolutionize the entire customer operations function,
improving the customer experience and agent productivity through digital self-service
and enhancing and augmenting agent skills. The technology has already gained traction
in customer service because of its ability to automate interactions with customers using
natural language. Research found that at one company with 5,000 customer service
agents, the application of generative AI increased issue resolution by 14 percent an hour and
reduced the time spent handling an issue by 9 percent.⁷ It also reduced agent attrition and
requests to speak to a manager by 25 percent. Crucially, productivity and quality of service
improved most among less-experienced agents, while the AI assistant did not increase—
and sometimes decreased—the productivity and quality metrics of more highly skilled
agents. This is because AI assistance helped less-experienced agents communicate using
techniques similar to those of their higher-skilled counterparts.

The following are examples of the operational improvements generative AI can have for
specific use cases:

— Customer self-service. Generative AI–fueled chatbots can give immediate and
personalized responses to complex customer inquiries regardless of the language or
location of the customer. By improving the quality and effectiveness of interactions via
automated channels, generative AI could automate responses to a higher percentage of
customer inquiries, enabling customer care teams to take on inquiries that can only be
resolved by a human agent. Our research found that roughly half of customer contacts
made by banking, telecommunications, and utilities companies in North America are
already handled by machines, including but not exclusively AI. We estimate that generative
AI could further reduce the volume of human-serviced contacts by up to 50 percent,
depending on a company’s existing level of automation.

— Resolution during initial contact. Generative AI can instantly retrieve data a company
has on a specific customer, which can help a human customer service representative more
successfully answer questions and resolve issues during an initial interaction.

— Reduced response time. Generative AI can cut the time a human sales representative
spends responding to a customer by providing assistance in real time and recommending
next steps.

— Increased sales. Because of its ability to rapidly process data on customers and their
browsing histories, the technology can identify product suggestions and deals tailored
to customer preferences. Additionally, generative AI can enhance quality assurance and
coaching by gathering insights from customer conversations, determining what could be
done better, and coaching agents.

We estimate that applying generative AI to customer care functions could increase
productivity at a value ranging from 30 to 45 percent of current function costs.

Our analysis captures only the direct impact generative AI might have on the productivity of
customer operations. It does not account for potential knock-on effects the technology may
have on customer satisfaction and retention arising from an improved experience, including
better understanding of the customer’s context that can assist human agents in providing
more personalized help and recommendations.


-----

_Marketing and sales_
Generative AI has taken hold rapidly in marketing and sales functions, in which text-based
communications and personalization at scale are driving forces. The technology can create
personalized messages tailored to individual customer interests, preferences, and behaviors,
as well as do tasks such as producing first drafts of brand advertising, headlines, slogans,
social media posts, and product descriptions.

However, introducing generative AI to marketing functions requires careful consideration.
For one thing, using mathematical models trained on publicly available data without
sufficient safeguards against plagiarism, copyright violations, and branding recognition risks
infringing on intellectual property rights. A virtual try-on application may produce biased
representations of certain demographics because of limited or biased training data. Thus,
significant human oversight is required for conceptual and strategic thinking specific to each
company’s needs.

Potential operational benefits from using generative AI for marketing include the following:

— Efficient and effective content creation. Generative AI could significantly reduce the
time required for ideation and content drafting, saving valuable time and effort. It can also
facilitate consistency across different pieces of content, ensuring a uniform brand voice,
writing style, and format. Team members can collaborate via generative AI, which can
integrate their ideas into a single cohesive piece. This would allow teams to significantly
enhance personalization of marketing messages aimed at different customer segments,
geographies, and demographics. Mass email campaigns can be instantly translated into
as many languages as needed, with different imagery and messaging depending on the
audience. Generative AI’s ability to produce content with varying specifications could
increase customer value, attraction, conversion, and retention over a lifetime and at a
scale beyond what is currently possible through traditional techniques.

— Enhanced use of data. Generative AI could help marketing functions overcome the
challenges of unstructured, inconsistent, and disconnected data—for example, from
different databases—by interpreting abstract data sources such as text, image, and
varying structures. It can help marketers better use data such as territory performance,
synthesized customer feedback, and customer behavior to generate data-informed
marketing strategies such as targeted customer profiles and channel recommendations.
Such tools could identify and synthesize trends, key drivers, and market and product
opportunities from unstructured data such as social media, news, academic research, and
customer feedback.

— SEO optimization. Generative AI can help marketers achieve higher conversion and
lower cost through search engine optimization (SEO) for marketing and sales technical
components such as page titles, image tags, and URLs. It can synthesize key SEO tokens,
support specialists in SEO digital content creation, and distribute targeted content to
customers.

— Product discovery and search personalization. With generative AI, product discovery
and search can be personalized with multimodal inputs from text, images and speech, and
deep understanding of customer profiles. For example, technology can leverage individual
user preferences, behavior, and purchase history to help customers discover the most


-----

relevant products and generate personalized product descriptions. This would allow
CPG, travel, and retail companies to improve their e-commerce sales by achieving higher
website conversion rates.

We estimate that generative AI could increase the productivity of the marketing function with
a value between 5 and 15 percent of total marketing spending.

Our analysis of the potential use of generative AI in marketing doesn’t account for knock-on
effects beyond the direct impacts on productivity. Generative AI–enabled synthesis could
provide higher-quality data insights, leading to new ideas for marketing campaigns and
better-targeted customer segments. Marketing functions could shift resources to producing
higher-quality content for owned channels, potentially reducing spending on external
channels and agencies.

Generative AI could also change the way both B2B and B2C companies approach sales. The
following are two use cases for sales:

— Increase probability of sale. Generative AI could identify and prioritize sales leads
by creating comprehensive consumer profiles from structured and unstructured data
and suggesting actions to staff to improve client engagement at every point of contact.
For example, generative AI could provide better information about client preferences,
potentially improving close rates.

— Improve lead development. Generative AI could help sales representatives nurture leads
by synthesizing relevant product sales information and customer profiles and creating
discussion scripts to facilitate customer conversation, including up- and cross-selling
talking points. It could also automate sales follow-ups and passively nurture leads until
clients are ready for direct interaction with a human sales agent.

Our analysis suggests that implementing generative AI could increase sales productivity by
approximately 3 to 5 percent of current global sales expenditures.

This analysis may not fully account for additional revenue that generative AI could bring
to sales functions. For instance, generative AI’s ability to identify leads and follow-up
capabilities could uncover new leads and facilitate more effective outreach that would bring
in additional revenue. Also, the time saved by sales representatives due to generative AI’s
capabilities could be invested in higher-quality customer interactions, resulting in increased
sales success.

Generative AI as a virtual collaborator
In other cases, generative AI can drive value by working in partnership with workers,
augmenting their work in ways that accelerate their productivity. Its ability to rapidly digest
mountains of data and draw conclusions from it enables the technology to offer insights and
options that can dramatically enhance knowledge work. This can significantly speed up the
process of developing a product and allow employees to devote more time to higher-impact
tasks.


-----

**Generative AI could increase**
**sales productivity by 3 to**
**5 percent of current global**
**sales expenditures.**

_Software engineering_
Treating computer languages as just another language opens new possibilities for software
engineering. Software engineers can use generative AI in pair programming and to do
augmented coding and train LLMs to develop applications that generate code when given a
natural-language prompt describing what that code should do.

Software engineering is a significant function in most companies, and it continues to grow
as all large companies, not just tech titans, embed software in a wide array of products and
services. For example, much of the value of new vehicles comes from digital features such as
adaptive cruise control, parking assistance, and IoT connectivity.

According to our analysis, the direct impact of AI on the productivity of software engineering
could range from 20 to 45 percent of current annual spending on the function. This value
would arise primarily from reducing time spent on certain activities, such as generating initial
code drafts, code correction and refactoring, root-cause analysis, and generating new system
designs. By accelerating the coding process, generative AI could push the skill sets and
capabilities needed in software engineering toward code and architecture design. One study
found that software developers using Microsoft’s GitHub Copilot completed tasks 56 percent
faster than those not using the tool.⁸ An internal McKinsey empirical study of software
engineering teams found those who were trained to use generative AI tools rapidly reduced
the time needed to generate and refactor code—and engineers also reported a better work
experience, citing improvements in happiness, flow, and fulfillment.

Our analysis did not account for the increase in application quality and the resulting boost in
productivity that generative AI could bring by improving code or enhancing IT architecture—
which can improve productivity across the IT value chain. However, the quality of IT
architecture still largely depends on software architects, rather than on initial drafts that
generative AI’s current capabilities allow it to produce.

Large technology companies are already selling generative AI for software engineering,
including GitHub Copilot, which is now integrated with OpenAI’s GPT-4, and Replit, used by
more than 20 million coders.⁹


-----

_Product R&D_
Generative AI’s potential in R&D is perhaps less well recognized than its potential in other business
functions. Still, our research indicates the technology could deliver productivity with a value ranging
from 10 to 15 percent of overall R&D costs.

For example, the life sciences and chemical industries have begun using generative AI foundation
models in their R&D for what is known as generative design. Foundation models can generate
candidate molecules, accelerating the process of developing new drugs and materials. Entos, a
biotech pharmaceutical company, has paired generative AI with automated synthetic development
tools to design small-molecule therapeutics. But the same principles can be applied to the design of
many other products, including larger-scale physical products and electrical circuits, among others.

While other generative design techniques have already unlocked some of the potential to apply AI
in R&D, their cost and data requirements, such as the use of “traditional” machine learning, can limit
their application. Pretrained foundation models that underpin generative AI, or models that have
been enhanced with fine-tuning, have much broader areas of application than models optimized for
a single task. They can therefore accelerate time to market and broaden the types of products to
which generative design can be applied. For now, however, foundation models lack the capabilities
to help design products across all industries.

In addition to the productivity gains that result from being able to quickly produce candidate
designs, generative design can also enable improvements in the designs themselves, as in the
following examples of the operational improvements generative AI could bring:

— Enhanced design. Generative AI can help product designers reduce costs by selecting and
using materials more efficiently. It can also optimize designs for manufacturing, which can lead to
cost reductions in logistics and production.

— Improved product testing and quality. Using generative AI in generative design can produce
a higher-quality product, resulting in increased attractiveness and market appeal. Generative
AI can help to reduce testing time of complex systems and accelerate trial phases involving
customer testing through its ability to draft scenarios and profile testing candidates.

We also identified a new R&D use case for nongenerative AI: deep learning surrogates, the use
of which has grown since our earlier research, can be paired with generative AI to produce even
greater benefits (see Box 2, “Deep learning surrogates”). To be sure, integration will require the
development of specific solutions, but the value could be significant because deep learning
surrogates have the potential to accelerate the testing of designs proposed by generative AI.

While we have estimated the potential direct impacts of generative AI on the R&D function, we
did not attempt to estimate the technology’s potential to create entirely novel product categories.
These are the types of innovations that can produce step changes not only in the performance of
individual companies but in economic growth overall.

**Value potential by modality**

Technology has revolutionized the way we conduct business, and text-based AI is on the frontier
of this change. Indeed, text-based data is plentiful, accessible, and easily processed and analyzed
at large scale by LLMs, which has prompted a strong emphasis on them in the initial stages of
generative AI development. The current investment landscape in generative AI is also heavily
focused on text-based applications such as chatbots, virtual assistants, and language translation.
However, we estimate that almost one-fifth of the value that generative AI can unlock across our use
cases would take advantage of multimodal capabilities beyond text to text.


-----

Box 2
**Deep learning surrogates**

Product design in industries producing
physical products often involves physicsbased virtual simulations such as
computational fluid dynamics (CFD) and
finite element analysis (FEA). Although
they are faster than actual physical
testing, these techniques can be timeand resource-intensive, especially for
designing complex parts—running CFD
simulations on graphics processing units


can take hours. And these techniques
are even more complex and computeintensive when they involve simulations
coupled across multiple disciplines (for
example, physical stress and temperature
distribution), which is sometimes called
multiphysics.

Deep learning applications are now
revolutionizing the virtual testing phase of


the R&D process by using deep learning
models to emulate (multi)physicsbased simulations at higher speeds and
lower costs. Instead of taking hours
to run physics-based models, these
deep learning surrogates can produce
the results of simulations in just a few
seconds, allowing researchers to test
many more designs and enabling faster
decision making on products and designs.


While most of generative AI’s initial traction has been in text-based use cases, recent advances in
generative AI have also led to breakthroughs in image generation, as OpenAI’s DALL·E and Stable
Diffusion have so amply illustrated, and much progress is being made in audio, including voice
and music, and video. These capabilities have obvious applications in marketing for generating
advertising materials and other marketing content, and these technologies are already being applied
in media industries, including game design. Indeed, some of these examples challenge existing
business models around talent, monetization, and intellectual property.[10]

The multimodal capabilities of generative AI could also be used effectively in R&D. Generative AI
systems could create first drafts of circuit designs, architectural drawings, structural engineering
designs, and thermal designs based on prompts that describe requirements for a product.
Achieving this will require training foundation models in these domains (think of LLMs trained on
“design languages”). Once trained, such foundation models could increase productivity on a similar
magnitude to software development.

**Value potential by industry**

Across the 63 use cases we analyzed, generative AI has the potential to generate $2.6 trillion to $4.4
trillion in value across industries. Its precise impact will depend on a variety of factors, such as the
mix and importance of different functions, as well as the scale of an industry’s revenue (Exhibit 4).


-----

Exhibit 4

Generative AI use cases will have diferent impacts on business functions
across industries.












|Generative AI productivity impact by business functions¹ C S Supply c S Tale|Col2|Col3|Col4|Col5|Col6|Col7|Col8|Col9|Col10|
|---|---|---|---|---|---|---|---|---|---|
|Low impact High impact M sales nd ap o g inr te kem rto as Cu s nDg in io& tr R e a t re c en i u dgp no ro e d P ren a a in tw fa oh c S sl na iog le t a rd en a d k sn ia y R Strateg e2 cT I n ae nt ra f rpo rg o Co and lent anization Total, % of industry Total, 760– 340– 230– 580– 290– 180– 120– 40– 60– revenue $ billion 1,200 470 420 1,200 550 260 260 50 90 Administrative and professional services 0.9–1.4 150–250||||||||||
|||||||||||
|Advanced electronics 1.3–2.3 100–170 and semiconductors||||||||||
|Advanced manufacturing3 1.4–2.4 170–290||||||||||
|Agriculture 0.6–1.0 40–70||||||||||
|Banking 2.8–4.7 200–340||||||||||
|Basic materials 0.7–1.2 120–200||||||||||
|Chemical 0.8–1.3 80–140||||||||||
|Construction 0.7–1.2 90–150||||||||||
|Consumer packaged goods 1.4–2.3 160–270||||||||||
|Education 2.2–4.0 120–230||||||||||
|Energy 1.0–1.6 150–240||||||||||
|Healthcare 1.8–3.2 150–260||||||||||
|High tech 4.8–9.3 240–460||||||||||
|Insurance 1.8–2.8 50–70||||||||||
|Media and entertainment 1.8–3.1 80–130||||||||||
|Pharmaceuticals and 2.6–4.5 60–110 medical products||||||||||
|Public and social sector 0.5–0.9 70–110||||||||||
|Real estate 1.0–1.7 110–180||||||||||
|Retail4 1.2–1.9 240–390||||||||||
|Telecommunications 2.3–3.7 60–100||||||||||
|Travel, transport, and logistics 1.2–2.0 180–300||||||||||


2,600–4,400

Note: Figures may not sum to 100%, because of rounding.
1Excludes implementation costs (eg, training, licenses).
2Excluding software engineering.
3Includes aerospace, defense, and auto manufacturing.
4Including auto retail.
Source: Comparative Industry Service (CIS), IHS Markit; Oxford Economics; McKinsey Corporate and Business Functions database; McKinsey Manufacturing
and Supply Chain 360; McKinsey Sales Navigator; Ignite, a McKinsey database; McKinsey analysis


McKinsey & Company


-----

For example, our analysis estimates generative AI could contribute roughly $310 billion in
additional value for the retail industry (including auto dealerships) by boosting performance in
functions such as marketing and customer interactions. By comparison, the bulk of potential
value in high tech comes from generative AI’s ability to increase the speed and efficiency of
software development.


-----

## 3


**The generative AI future**
**of work: Impacts on**
**work activities, economic**
**growth, and productivity**

Technology has been changing the anatomy of work for decades. Over the years, machines
have given human workers various “superpowers”; for instance, industrial-age machines
enabled workers to accomplish physical tasks beyond the capabilities of their own bodies.
More recently, computers have enabled knowledge workers to perform calculations that
would have taken years to do manually.

These examples illustrate how technology can augment work through the automation of
individual activities that workers would have otherwise had to do themselves. At a conceptual
level, the application of generative AI may follow the same pattern in the modern workplace,
although as we show later in this chapter, the types of activities that generative AI could
affect, and the types of occupations with activities that could change, will likely be different as
a result of this technology than for older technologies.


-----

The McKinsey Global Institute began analyzing the impact of technological automation of
work activities and modeling scenarios of adoption in 2017. At that time, we estimated that
workers spent half of their time on activities that had the potential to be automated by adapting
technology that existed at that time, or what we call technical automation potential. We also
modeled a range of potential scenarios for the pace at which these technologies could be
adopted and affect work activities throughout the global economy.

Technology adoption at scale does not occur overnight. The potential of technological
capabilities in a lab does not necessarily mean they can be immediately integrated into a
solution that automates a specific work activity—developing such solutions takes time. Even
when such a solution is developed, it might not be economically feasible to use if its costs
exceed those of human labor. Additionally, even if economic incentives for deployment exist, it
takes time for adoption to spread across the global economy. Hence, our adoption scenarios,
which consider these factors together with the technical automation potential, provide a sense
of the pace and scale at which workers’ activities could shift over time.

Large-scale shifts in the mix of work activities and occupations are not unprecedented.
Consider the work of a farmer today compared with what a farmer did just a few short years
ago. Many farmers now access market information on mobile phones to determine when and
where to sell their crops or download sophisticated modeling of weather patterns. From a more
macro perspective, agricultural employment in China went from an 82 percent share of all
workers in 1962 to 13 percent in 2013. Labor markets are also dynamic: millions of people leave
their jobs every month in the United States[.11] But this does not minimize the challenges faced
by individual workers whose lives are upended by these shifts, or the organizational or societal
challenges of ensuring that workers have the skills to take on the work that will be in demand
and that their incomes are sufficient to grow their standards of living.

Also, demographics have made such shifts in activities a necessity from a macroeconomic
perspective. An economic growth gap has opened as a result of the slowing growth of the
world’s workforce. In some major countries, workforces have shrunk because populations are
aging. Labor productivity will have to accelerate to achieve economic growth and enhance
prosperity.

The analyses in this paper incorporate the potential impact of generative AI on today’s work
activities. The new capabilities of generative AI, combined with previous technologies and
integrated into corporate operations around the world, could accelerate the potential for
technical automation of individual activities and the adoption of technologies that augment the
capabilities of the workforce. They could also have an impact on knowledge workers whose
activities were not expected to shift as a result of these technologies until later in the future.


-----

**Accelerating the technical potential to transform knowledge work**

Based on developments in generative AI, technology performance is now expected to
match median human performance and reach top-quartile human performance earlier
than previously estimated across a wide range of capabilities (Exhibit 5). For example, MGI
previously identifi ed 2027 as the earliest year when median human performance for naturallanguage understanding might be achieved in technology, but in this new analysis, the
corresponding point is 2023.

Exhibit 5

As a result of generative AI, experts assess that technology could achieve humanlevel performance in some technical capabilities sooner than previously thought.

Technical capabilities, level of human performance achievable by technology


Estimates post-recent

of expert estimates

Estimates pre-generative AI (2017)¹ Median Top quartile


Coordination with multiple agents

Creativity

Logical reasoning and problem solving

Natural-language generation

Natural-language understanding

Output articulation and presentation

Generating novel patterns and categories

Sensory perception

Social and emotional output

Social and emotional reasoning

Social and emotional sensing

¹Comparison made on the business-related tasks required from human workers. Please refer to technical appendix for detailed view of performance
rating methodology.
Source: McKinsey Global Institute occupation database; McKinsey analysis

McKinsey & Company


-----

As a result of these reassessments of technology capabilities due to generative AI, the total
percentage of hours that could theoretically be automated by integrating technologies that exist
today has increased from about 50 percent to 60 to 70 percent. The technical potential curve is
quite steep because of the acceleration in generative AI’s natural-language capabilities (Exhibit 6).

Interestingly, the range of times between the early and late scenarios has compressed compared
with the expert assessments in 2017, refl ecting a greater confi dence that higher levels of
technological capabilities will arrive by certain time periods.

Exhibit 6

The advent of generative AI has pulled forward the potential for
technical automation.

Technical automation potentials by scenario, %

100
Updated early scenario
including generative AI[2]


Updated late scenario
including generative AI[2]

2017 early scenario[2]

2017 late scenario[2]


90

80

70

60

50


Time spent on
current work
activities1


|Col1|20|Col3|23|Col5|Col6|Col7|Col8|
|---|---|---|---|---|---|---|---|
|||||||||
|||||||||
|||||||||
|||||||||
|||||||||


2020 2030 2040 2050 2060


1Includes data from 47 countries, representing about 80% of employment across the world. 2017 estimates are based on the activity and occupation mix from
2016. Scenarios including generative AI are based on the 2021 activity and occupation mix.
2Early and late scenarios refect the ranges provided by experts (see Exhibit 6).
Source: McKinsey Global Institute analysis

McKinsey & Company

**Generative AI could propel higher productivity growth**

Global economic growth was slower from 2012 to 2022 than in the two preceding decades.[12]

Although the COVID-19 pandemic was a signifi cant factor, long-term structural challenges—
including declining birth rates and aging populations—are ongoing obstacles to growth.

Declining employment is among those obstacles. Compound annual growth in the total number
of workers worldwide slowed from 2.5 percent in 1972–82 to just 0.8 percent in 2012–22,
largely because of aging. In many large countries, the size of the workforce is already declining.[13]


-----

Productivity, which measures output relative to input, or the value of goods and services
produced divided by the amount of labor, capital, and other resources required to produce
them, was the main engine of economic growth in the three decades from 1992 to 2022
(Exhibit 7). However, since then, productivity growth has slowed in tandem with slowing
employment growth, confounding economists and policy makers.[14]

Exhibit 7

Productivity growth, the main engine of GDP growth over the past 30 years,
slowed down in the past decade.


Real GDP growth contribution of employment
and productivity growth, 1972–2022,
global GDP growth, CAGR, %

Productivity growth 0.7



_Productivity growth bigger contributor_
_to GDP growth_

3.8

3.1 3.1 1.3
2.9

2.8

0.8

1.4

2.5
2.0

2.5
2.1

1.7

0.8
0.7


1972–82 1982–92 1992–2002 2002–12 2012–22

Source: Conference Board Total Economy database; McKinsey Global Institute analysis

McKinsey & Company

The deployment of generative AI and other technologies could help accelerate productivity
growth, partially compensating for declining employment growth and enabling overall
economic growth. Based on our estimates, the automation of individual work activities
enabled by these technologies could provide the global economy with an annual productivity
boost of 0.5 to 3.4 percent from 2023 to 2040 depending on the rate of automation
adoption—with generative AI contributing to 0.1 to 0.6 percentage points of that growth—but
only if individuals aff ected by the technology were to shift to other work activities that at
least match their 2022 productivity levels (Exhibit 8). In some cases, workers will stay in the
same occupations, but their mix of activities will shift; in others, workers will need to shift
occupations.


-----

Exhibit 8

Generative AI could contribute to productivity growth if labor hours can
be redeployed efectively.

Productivity impact from automation by scenario, 2022–40, CAGR,¹ %

Without generative AI² Additional with generative AI

Japan Germany France United States


3.8

0.6

3.2

1.1

0.8

Early Late


3.8

0.7

3.1

1.0

0.7

Early Late


0.2


0.3

0.1


0.2

3.7

0.6

3.1

0.8

0.6


Early Late

Emerging economies


0.2


3.7

0.7

3.0

0.8

0.6


Early Late


Global³

3.4

0.6

2.8

0.5

0.1

0.3

Early Late


China Mexico India South Africa


0.1


0.0


3.6

0.6

3.0

0.5

0.4

Early Late


3.4

0.5

2.8

0.3

0.3

Early Late


0.1


3.1

0.4

2.7

0.1


2.9
0.5

2.4

0.1


Early Late


Early Late


Note: Figures may not sum, because of rounding.
1Based on the assumption that automated work hours are reintegrated in work at productivity level of today.
2Previous assessment of work automation before the rise of generative AI.
3Based on 47 countries, representing about 80% of world employment.
Source: Conference Board Total Economy Database; Oxford Economics; McKinsey Global Institute analysis

McKinsey & Company

The capabilities of generative AI vastly expand the pool of work activities with the potential for
technical automation. That in turn has sped up the pace at which automation may be deployed and
expanded the types of workers who will experience its impact. Like other technologies, its ability
to take on routine tasks and work can increase human productivity, which has grown at a belowaverage rate for almost 20 years.[15] It can also off set the impact of aging, which is beginning to put a
dent in workforce growth for many of the world’s major economies. But to achieve these benefi ts,
a signifi cant number of workers will need to substantially change the work they do, either in their
existing occupations or in new ones. They will also need support in making transitions to new
activities.


-----

## 4


**Considerations for**
**businesses and society**

History has shown that new technologies have the potential to reshape societies. Artificial
intelligence has already changed the way we live and work—for example, it can help our phones
(mostly) understand what we say, or draft emails. Mostly, however, AI has remained behind the
scenes, optimizing business processes or making recommendations about the next product to buy.
The rapid development of generative AI is likely to significantly augment the impact of AI overall,
generating trillions of dollars of additional value each year and transforming the nature of work.

But the technology could also deliver new and significant challenges. Stakeholders must act—and
quickly, given the pace at which generative AI could be adopted—to prepare to address both the
opportunities and the risks. Risks have already surfaced, including concerns about the content that
generative AI systems produce: Will they infringe upon intellectual property due to “plagiarism”
in the training data used to create foundation models? Will the answers that LLMs produce when
questioned be accurate, and can they be explained? Will the content that generative AI creates
be fair or biased in ways that users do not want by, say, producing content that reflects harmful
stereotypes?

There are economic challenges too: the scale and the scope of the workforce transitions described
in this report are considerable. In the midpoint adoption scenario, about a quarter to a third of
work activities could change in the coming decade. The task before us is to manage the potential


-----

positives and negatives of the technology simultaneously (for more about the potential risks of
generative AI, see Box 3, “Using generative AI responsibly”). Here are some of the critical questions
we will need to address while balancing our enthusiasm for the potential benefits of the technology
with the new challenges it can introduce.

Companies and business leaders
How can companies move quickly to capture the potential value at stake highlighted in this report,
while managing the risks that generative AI presents?


Box 3

**Using generative AI responsibly**

Generative AI poses a variety of risks.
Stakeholders will want to address these
risks from the start.

Fairness: Models may generate
algorithmic bias due to imperfect training
data or decisions made by the engineers
developing the models.

Intellectual property (IP): Training
data and model outputs can generate
significant IP risks, including infringing
on copyrighted, trademarked, patented,
or otherwise legally protected materials.
Even when using a provider’s generative
AI tool, organizations will need to
understand what data went into training
and how it’s used in tool outputs.

Privacy: Privacy concerns could arise if
users input information that later ends
up in model outputs in a form that makes


individuals identifiable. Generative
AI could also be used to create and
disseminate malicious content such as
disinformation, deepfakes, and hate
speech.

Security: Generative AI may be
used by bad actors to accelerate the
sophistication and speed of cyberattacks.
It also can be manipulated to provide
malicious outputs. For example, through a
technique called prompt injection, a third
party gives a model new instructions that
trick the model into delivering an output
unintended by the model producer and
end user.

Explainability: Generative AI relies
on neural networks with billions of
parameters, challenging our ability


to explain how any given answer is
produced.

Reliability: Models can produce different
answers to the same prompts, impeding
the user’s ability to assess the accuracy
and reliability of outputs.

Organizational impact: Generative AI
may significantly affect the workforce,
and the impact on specific groups
and local communities could be
disproportionately negative.

Social and environmental impact: The
development and training of foundation
models may lead to detrimental social and
environmental consequences, including
an increase in carbon emissions (for
example, training one large language
model can emit about 315 tons of carbon
dioxide).[1]


1 Ananya Ganesh, Andrew McCallum, and Emma Strubell, “Energy and policy considerations for deep learning in NLP,” Proceedings of the 57th Annual Meeting of the
_Association for Computational Linguistics, June 5, 2019._

How will the mix of occupations and skills needed across a company’s workforce be transformed
by generative AI and other artificial intelligence over the coming years? How will a company enable
these transitions in its hiring plans, retraining programs, and other aspects of human resources?

Do companies have a role to play in ensuring the technology is not deployed in “negative use cases”
that could harm society?

How can businesses transparently share their experiences with scaling the use of generative AI
within and across industries—and also with governments and society?


-----

Policy makers
What will the future of work look like at the level of an economy in terms of occupations and skills?
What does this mean for workforce planning?

How can workers be supported as their activities shift over time? What retraining programs can be put
in place? What incentives are needed to support private companies as they invest in human capital?
Are there earn-while-you-learn programs such as apprenticeships that could enable people to retrain
while continuing to support themselves and their families?

What steps can policy makers take to prevent generative AI from being used in ways that harm society
or vulnerable populations?

Can new policies be developed and existing policies amended to ensure human-centric AI
development and deployment that includes human oversight and diverse perspectives and accounts
for societal values?

Individuals as workers, consumers, and citizens
How concerned should individuals be about the advent of generative AI? While companies can assess
how the technology will affect their bottom lines, where can citizens turn for accurate, unbiased
information about how it will affect their lives and livelihoods?

How can individuals as workers and consumers balance the conveniences generative AI delivers with
its impact in their workplaces?

Can citizens have a voice in the decisions that will shape the deployment and integration of generative
AI into the fabric of their lives?

Technological innovation can inspire equal parts awe and concern. When that innovation seems
to materialize fully formed and becomes widespread seemingly overnight, both responses can
be amplified. The arrival of generative AI in the fall of 2022 was the most recent example of this
phenomenon, due to its unexpectedly rapid adoption as well as the ensuing scramble among
companies and consumers to deploy, integrate, and play with it.

All of us are at the beginning of a journey to understand this technology’s power, reach, and
capabilities. If the past eight months are any guide, the next several years will take us on a rollercoaster ride featuring fast-paced innovation and technological breakthroughs that force us to
recalibrate our understanding of AI’s impact on our work and our lives. It is important to properly
understand this phenomenon and anticipate its impact. Given the speed of generative AI’s deployment
so far, the need to accelerate digital transformation and reskill labor forces is great.

These tools have the potential to create enormous value for the global economy at a time when it is
pondering the huge costs of adapting to and mitigating climate change. At the same time, they also
have the potential to be more destabilizing than previous generations of artificial intelligence. They are
capable of that most human of abilities, language, which is a fundamental requirement of most work
activities linked to expertise and knowledge as well as a skill that can be used to hurt feelings, create
misunderstandings, obscure truth, and incite violence and even wars.

We hope this research has contributed to a better understanding of generative AI’s capacity to
add value to company operations and fuel economic growth and prosperity as well as its potential
to dramatically transform how we work and our purpose in society. Companies, policy makers,
consumers, and citizens can work together to ensure that generative AI delivers on its promise to
create significant value while limiting its potential to upset lives and livelihoods. The time to act is now.[16]


-----

**Endnotes**

1 Ryan Morrison, “Compute power is becoming
a bottleneck for developing AI. Here’s how
you clear it.,” Tech Monitor, updated March 17,
2023.

2 “Introducing ChatGPT,” OpenAI, November
30, 2022; “GPT-4 is OpenAI’s most advanced
system, producing safer and more useful
responses,” OpenAI, accessed June 1, 2023.

3 “Introducing Claude,” Anthropic PBC,
March 14, 2023; “Introducing 100K Context
Windows,” Anthropic PBC, May 11, 2023.

4 Emma Roth, “The nine biggest announcements
from Google I/O 2023,” The Verge, May 10,
2023.

5 Pitchbook.

6 Ibid.

7 Erik Brynjolfsson, Danielle Li, and Lindsey
R. Raymond, Generative AI at work, National
Bureau of Economic Research working paper
number 31161, April 2023.

8 Peter Cihon et al., The impact of AI on
_developer productivity: Evidence from GitHub_
_Copilot, Cornell University arXiv software_
engineering working paper, arXiv:2302.06590,
February 13, 2023.


9 Michael Nuñez, “Google and Replit join forces
to challenge Microsoft in coding tools,”
VentureBeat, March 28, 2023.

10 Joe Coscarelli, “An A.I. hit of fake ‘Drake’ and
‘The Weeknd’ rattles the music world,” New
York Times, updated April 24, 2023.

11 “Job openings and labor turnover survey,” US
Bureau of Labor Statistics, accessed June 6,
2023.

12 _Global economic prospects, World Bank,_
January 2023.

13 Yaron Shamir, “Three factors contributing to
fewer people in the workforce,” Forbes, April 7,
2022.

14 “The U.S. productivity slowdown: an economywide and industry-level analysis,” Monthly
Labor Review, US Bureau of Labor Statistics,
April 2021; Kweilin Ellingrud, “Turning around
the productivity slowdown,” McKinsey Global
Institute, September 13, 2022.

15 “Rekindling US productivity for a new era,”
McKinsey Global Institute, February 16, 2023.

16 The research, analysis, and writing in this
report was entirely done by humans.


The research underpinning this report was led by Michael Chui, an MGI partner in McKinsey’s Bay Area office; Eric Hazan, a
senior partner in the Paris office; Roger Roberts, a partner in the Bay Area office; Alex Singla, a senior partner in the Chicago
office; Kate Smaje and Alexander Sukharevsky, senior partners in the London office; Lareina Yee, a senior partner in the
Bay Area office; and Rodney Zemmel, a senior partner in the New York office.


-----

##### Making the most of the generative AI opportunity: Six questions for CEOs

As corporate leaders navigate the new gen AI era, they can begin to lay out
their road map and strategy by pondering a series of fundamental questions.

_This article is a collaborative effort by Ben Ellencweig, Dana Maor, Alex Singla, Alexander Sukharevsky,_
_Lareina Yee, and Rodney Zemmel, representing views from QuantumBlack, AI by McKinsey._

© Getty Images


-----

Generative AI (gen AI) has taken the world
by storm, altering our understanding of the
possible. Creating next-era fashion collections
in a few clicks, engaging customers with hyperpersonalized offerings, and collapsing years of
tedious drug discovery work into a few months—
suddenly, all that and more seems within reach.
As in the early days of breakthroughs like
blockchain and the Internet itself, gen AI has
sparked a debate between those who believe
the technology will reshape the way we work and
live and those who see gen AI as the next NFT
moment, soaring briefly and failing to deliver on
its promise, as nonfungible tokens did earlier in
this decade.

So how much of today’s excitement about gen AI
reflects reality, and how much is myth? McKinsey
estimates that the technology will open a new era
of productivity and growth that could create $2.6
billion to $4.4 trillion of additional value.¹ In the
telecom space alone, the impact of new gen AI
use cases is expected to be in the range of $60
billion to $100 billion.

For CEOs seeking to unlock this upside, the key
is to understand how this value will materialize
and over what period, as well as where to invest
their resources. There are no right answers, at
least not yet. We are still in the technology’s
post-awareness, pre-deployment phase, with
most software engineers having only recently
gained access to gen AI tools. But based on our
experience working with clients over the past 15
months, we find that CEOs can better formulate
a strategy if they consider six essential questions
about gen AI:

1. Is the opportunity significantly larger than AI?

2. Are we ambitious enough with gen AI?

3. Where is the money in the value chain?

4. Do we have the right talent in place?


5. What does it take to cross the “Death Valley”
of scaling AI?

6. Are we thinking about risk in the right way?

**Is the opportunity significantly**
**larger than AI?**
Over the past year, many of our client
conversations and technology deployments have
focused on gen AI. Despite its novelty, however,
gen AI does not exist in a silo. Instead, it is simply
the newest, if most powerful, iteration in the
unfolding story of how artificial intelligence can
boost productivity and innovation. We estimate
that gen AI accounts for only 20 to 40 percent
of AI’s total value creation potential, with the
remainder coming from traditional, or “analytical,”
AI applications, which have heretofore been less
than fully deployed.

What’s more, other important technology trends,
such as Web 3.0 and augmented reality and
virtual reality (AR/VR), are continuing to make
progress in the shadow of gen AI. They will
eventually get a strong footing over the next
decade, with clear value creation potential for
organizations. Hence, executives rethinking
industries and business models should view the
opportunity more broadly than gen AI or even
all AI. A more effective approach is to consider
how their organizations can capitalize on the
confluence of emerging technology trends—a
truly watershed moment akin to the simultaneous
emergence of the first cloud, social network, and
smartphone applications in 2017.

**Are we ambitious enough with**
**gen AI?**
Gen AI has fascinated the world with jawdropping applications like ChatGPT and Pi,
highlighting AI’s transformative potential.
Never before has technology pushed the art


1 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.


-----

of the possible so far ahead and so fast for nontechnologists. As companies rush toward this
technology, they are likelier to succeed if they solve
for value creation versus simply checking the box.
This is particularly concerning in the telco space,
where players have often expressed interest in
exploring incremental productivity applications and
less frequently turn their attention to reimagining
their businesses through the lens of AI.

Beyond simply avoiding a rehashed discussion
of tech versus telco and scaling use cases, senior
executives can benefit from asking a weightier
question: How do we reinvent our industry and
business model by leveraging the disintermediation,
radical cost curve shifts, and organic consumer
acquisition opportunities that gen AI can provide?
Moreover, the age of new platforms opens new
opportunities, including the creation from scratch of
hyperscalers or unicorn super apps. One only needs
to consider the opportunity associated with naturallanguage virtual assistants and the disruption this
could have on the current business context, from
consumption to business model. Gen AI will reward
the bold. Already, some 80 percent of today’s
most popular gen AI products come from new
entrants,² with incumbents forced to play catch-up
or otherwise find their edge to lead.

**Where is the money in the value chain?**
Gen AI is creating a frenzy among founders and
investors, with a seemingly endless number of
players entering the field. A closer look at leading
gen AI players reveals a couple of winning plays that
CEOs might use to separate their organization from
the pack³ :

— Differentiate à la the fine-dining chef. The
ingredients of gen AI applications are not in
and of themselves a source of competitive
differentiation. Anyone can license the most
powerful closed-source models, which
contribute only about 15 percent of the value of


gen AI applications. This suggests that the real
value will be realized by those able to combine
the best available technology with proprietary
data. Telco leaders should reexamine their data
asset portfolios with an eye toward designing
features like unique consumer and distribution
journeys, such as an always-on customer care
assistant fine-tuned to each user profile and
embedded across user channels. Indeed, this
is exactly what people seem to be asking
for: among the top 50 gen AI applications,
consumers are paying for 90 percent of them,
revenue per user is three times higher than that
of other apps, and customer acquisition is mostly
organic.

— Find underserved segments of the value
chain. Gen AI models and applications get most
of the attention from investors and organizations,
but other critical segments of the gen AI value
chain remain surprisingly underrated. From
commercializing access to graphical processing
units (GPUs) to providing data cleaning,
augmentation, or risk management solutions,
the opportunities are plentiful. We see this play
already taking place in the data center space,
with investors exploring acquisitions to supply an
accelerating demand for workloads. Fortunately,
many opportunities still exist for organizations to
gain first-mover advantages in the gen AI market.
In fact, in most product categories, the gap
between the top two players is only two times,
making it easier for new entrants to establish
themselves as leaders in the field.

**Do we have the right talent in place?**
Our research shows that gen AI is expected to
supercharge automation, affecting up to 60 percent
of work activities over the next 20 years. This
impact should not be surprising; a gen AI model can
analyze in an hour more data than a human can in
ten lifetimes. But will AI replace us all or turn us into
automatons?


2 Olivia Moore, “How are consumers using generative AI?” Andreessen Horowitz, Sept. 13, 2023; By some estimates, gen AI start-ups alone have
already generated more than $1 billion of software-as-a-service revenue.
3 Ninety percent of these companies are already monetizing their offerings with more than three times the average revenue per user than
incumbents.


-----

Those concerns seem overwrought, at least for
now. The fact is, gen AI can deliver only if it is
combined with exceptional human capital. Despite
the power of gen AI, middling employees will
produce middling results. Organizations must
recruit, retain, and develop truly outstanding talent
in both the technical and nontechnical spheres.
With the right people in place, organizations truly
could be on the verge of a new age of innovation.

**What does it take to cross the ‘Death**
**Valley’ of scaling AI?**
Only one in ten AI use cases have been deployed
in production,⁴ so gen AI has arrived at a time
when many leaders are disillusioned with the
yet-unfulfilled promises of artificial intelligence.
But AI does not have a technology problem; it
has a design problem. To be effective, AI models
require top-down focus, the right tech and
people capabilities, proper data access, modular
architecture, and effective change management.
Only then can disparate AI-driven solutions work
together continuously to create great customer
and employee experiences, lower unit costs,
and allow the organization to move faster than
ever. Without external intervention or guidance,
only about 3 percent of gen AI proofs of concept
eventually scale.

Creating a digitally capable organization involves
rewiring the way companies operate. This effort
should be broad, covering six dimensions:

1. _business-led digital road map that aligns the_
senior leadership team on the transformation
vision, value, and strategy, which is focused on
business reinvention

2. _talent with the right skills and capabilities to_
execute and innovate in both the technical and
business sides of the organization, including
upskilling


3. _operating model that increases the_
organization’s metabolic rate by bringing
together business and technology

4. _technology that allows the organization to_
innovate faster and more easily—in particular,
an IT architecture with a flexible orchestration
layer

5. _data that is continuously enriched and easy_
to consume across the organization to
improve customer experiences and business
performance

6. _adoption and scaling of digital and AI solutions_
to optimize value capture by building new skills
and leadership characteristics and by tightly
managing the transformation progress and
risks

**Are we thinking about risk in the**
**right way?**
Discussions of gen AI risks are plentiful, but
experience shows that most of these conversations
need calibrating to ensure that organizations
approach risk holistically and pragmatically.
Current conversations about risk tend to focus
on either short-term considerations (for example,
customer experience and protection of intellectual
property) or long-term, existential ones (whether
artificial general intelligence will rule the world).
Not enough focus is placed on intermediate risk,
such as how companies can maintain the trust
of their stakeholders in an AI-generated reality
where seeing is no longer believing. Also, other
categories of risks are simply not getting the
attention they deserve. Little is said, for example,
about organizations’ environmental, social, and
governance (ESG) risk, even though training a gen
AI model consumes about a million liters of water
for cooling.


4 “The state of AI in 2022—and a half decade in review,” McKinsey, December 6, 2022.


-----

A more pragmatic perspective would be for CEOs
to steer their organizations toward accepting risk
as the reality of doing business with AI (for example,
hallucination is just a feature of gen AI). Fortunately,
risks can be managed. Plenty of banks, after all,
deal with customer credit and other difficult types
of risk daily and still manage to thrive. To navigate
these uncharted waters, organizations should set up
cross-functional teams to cover their specific risk
concerns (for example, regulatory, ethical, cyber, IP,
and societal risks), establish ethical principles and
guidelines for gen AI use, and establish continuous
monitoring for gen AI systems to address risk
dynamically.


An honest and thorough examination of these six
questions can lay the foundation of a comprehensive
gen AI strategy—one that truly focuses on how
the technology can transform an organization or
an entire industry. These conversations will not
necessarily be easy, which makes it essential
that they be led by CEOs. Perhaps most of all, it is
advantageous to think big. A road map, after all, can
lead to different destinations. Where do you want
your company to land?


Ben Ellencweig is a senior partner in McKinsey’s Stamford office, Dana Maor is a senior partner in the Tel Aviv office, and
Lareina Yee, a senior partner and chair of the McKinsey Technology Council, is based in the Bay Area–San Francisco office.
Alex Singla, a senior partner in the Chicago office, and Alexander Sukharevsky, a senior partner in the London office,
are managing partners of QuantumBlack, AI by McKinsey. Rodney Zemmel, a senior partner and managing partner of
McKinsey Digital, is based in the New York office.

Copyright © 2024 McKinsey & Company. All rights reserved.


-----

### Sector view: Telecom operators


-----

Technology, Media & Telecommunications Practice
**The AI-native telco: Radical**
**transformation to thrive in**
**turbulent times**

Artificial intelligence, when deployed at scale, can help telcos protect
core revenues and drive margin growth. But capturing this opportunity
will require a wholly different approach.

_This article is a collaborative effort by Joshan Abraham, Jorge Amar, Yuval Atsmon, Miguel Frade, and_
_Tomás Lajous, representing views from McKinsey’s Technology, Media & Telecommunications Practice._


© Getty Images


-----

Artificial intelligence (AI) is unlocking use cases
that are transforming industries across a wide
swath of the world’s economy. From infrastructure
that “self-heals” to radically reimagined (and
touchless) customer service and experience; from
large scale hyperpersonalization to automatically
created marketing messages and images leveraging
Generative AI tools like ChatGPT—it is all a reality
today. These AI solutions can powerfully augment
and sometimes radically outperform most traditional
business roles.

The impact from these solutions is becoming
evident. AI leaders—the top quintile of companies
that have taken the McKinsey Analytics Quotient
assessment—have experienced a five-year revenue
CAGR that is 2.1 times higher than that of peers and
a total return to shareholders that is 2.5 times larger.

Given the numerous challenges the telecom industry
has faced in recent years, such as flagging revenues
and ROIC, one might expect the industry would have
already adopted a full transition to this technology.
Yet, based on our experience with operators
across the world, telcos have yet to fully embrace
AI and an AI-focused mindset. Instead, models are
developed once and not enhanced as the business
context evolves. Machine learning (ML) is in name
only, limiting the ability of the system to improve
from experience. Most regrettably, AI investments
are often not aligned with top-level management
priorities; lacking that sponsorship, AI deployments
stall, investment in technical talent withers, and the
technology remains immature.

Contrast this disjointed state of affairs with an
AI-native organization. Here, AI is viewed as a
core competency that powers decision making
across all departments and organization layers.
AI investments are required to enable most
C-level priorities such as more personalized
recommendations for customers and faster speed
of answer in call centers. Top executives serve
as champions of critical AI initiatives. Data and AI
capabilities are managed as products, built for
scalability and reusability. AI product managers,


even those working on foundational products, are
celebrated for the benefits they generate for the
organization.

Reaching this state of AI maturity is no easy task,
but it is certainly within the reach of telcos. Indeed,
with all the pressures they face, embracing largescale deployment of AI and transitioning to being
AI-native organizations could be key to driving
growth and renewal. Telcos that are starting to
recognize this is nonnegotiable are scaling AI
investments as the business impact generated by
the technology materializes.

While isolated applications of the technology
can help individual departments improve, it’s AI
connected holistically at all levels and departments
that will be key to protecting core revenue and
driving margin growth in even the most difficult
of environments. Imagine the following not-sodistant scenarios:

— Customer focused: Sarah, a New Yorker, is
a high average revenue per user (ARPU)
customer. Aware that Sarah spends half of
her phone usage time on fitness apps, the AI
creates an enticing customized upgrade offer
that includes a six-month credit applicable
to her favorite fitness subscription and NYCspecific perks, such as a ticket to an upcoming
concert sponsored by the operator. Knowing
Sarah’s high digital propensity¹, the AI makes
the offer available to her as a digital-only
promotion.

— Employee focused: When Trevor, an associate
in a telco mall store, logs in at the start of his
shift, he receives a celebratory notification
congratulating him on his high-quality
interactions with customers the previous
day. And because the AI detected that Trevor
is underperforming peers in accessory and
device protection attach rates, he receives a
notification pointing him to coaching resources
specifically created to enhance performance in
those metrics.


1 Preference to transact and engage in digital channels, such as websites and mobile apps.


-----

— Infrastructure focused: Lucile, director of a
capital planning team, uses AI to inform highly
targeted network investment decisions based
on a granular understanding of customer-level
network experience scores strongly correlated to
commercial outcomes (for example, churn). The
AI provides tactical recommendations of what
and where to build based on where customers
use the network and on automatically computed
thresholds after which new investments have
marginal impact on experience and commercial
outcomes for the operator.

How these possibilities could become reality is critical
to consider, especially given that most telcos currently
deploy AI in limited ways that will not drive sustainable,
at-scale success.

**Why now? The case for becoming**
**AI native**
Factors supporting this move for telcos include the
following:

— Increasing accessibility of leading AI technology:
AI-native organizations like Meta continue to
grow the open-source ecosystem by making new
programming languages, datasets, and algorithms
widely available. In parallel, cloud providers have
developed multiple quick-to-deploy machinelearning APIs like Google Cloud’s Natural
Language API. Generative AI solutions, such as
ChatGPT, that are capable of creating engaging
responses to human queries are also accessible
through API. These two factors, coupled with
dropping costs of data processing and storage,
make AI increasingly easier for organizations to
leverage.

— Rapid explosion of usable data: Operators can
collect, structure, and use significantly more
data directly than ever before. This information
includes data flows from individualized app usage
patterns, site-specific customer experience
scores, and what can be purchased or shared
from partners or third parties. To answer privacy
fears raised by consumers and regulators, telcos
must also invest in building digital trust, including
actively managing data privacy and having a


robust cybersecurity strategy and a framework
to guide ethical deployment of AI.

— Proven use cases and outcomes: AI-native
organizations across industries have deployed AI
to achieve four critical outcomes highly relevant
to operators across the world: 1) drive revenue
protection and growth through personalization,
2) transform the cost structure, 3) enable a
frictionless customer experience, and 4) meet
new workplace demands. Operators can learn
from all of them. Streaming players, for example,
have long been known for providing highly
curated personalized content recommendations
based on past user behavior. To optimize cost
and deliver a seamless customer experience,
one of the leading US insurance companies
leverages AI assistants to reduce and even
eliminate human interactions for users to obtain
coverage or cancel policies with other carriers.
In turn, some of the leading tech companies in
the world are known for using AI to highlight the
traits of great managers and high-performing
teams and use those insights to train company
leaders.

— Technology investments recognized as a
_business driver: In a postpandemic world,_
there is broad consensus among investors and
executives that technology investments are not
a mere cost center but a fundamental business
driver with profound impacts on the bottom
line. Despite prospects of economic turmoil and
recessionary fears, IT spending is expected to
increase by more than 5 percent in 2023, with
technology leaders under growing pressure to
demonstrate impact on company financials.²

— Operator bets need hypercharging: As networks
and products converge, operators are making
bets on becoming cost and efficiency focused,
experience-centric, or ecosystem players. AI use
cases that are more relevant for each bet can
give them a better chance to hypercharge and
leapfrog competition.

For the greatest payoff, this shift requires
telcos to embrace the concept of the AI-native
organization—a structure where the technology


-----

is deeply embedded across the fabric of the
entire enterprise.

Using AI to reimagine the core business
Telcos have been under relentless pressure
over the past decade as traditional growth
drivers eroded and economic value increasingly
shifted to tech companies. By using AI to its
fullest extent, operators can protect their core
business from further erosion while improving
margins.

As the industry looks to leverage the power
of AI, we see six themes gaining prevalence in
strategic agendas based on our experience
working with telcos across the world.

Hyperpersonalize and architect sales and
engagement
Leveraging the breadth and depth of userlevel data at their disposal, operators have
been increasingly investing in AI-enabled
personalization and channel steering.

For example, a hyperpersonalized plan and
device recommendation for each line holder
could leverage granular behavioral data—such
as number of and engagement with apps
installed and device feature usage—to create
individualized plan recommendations (superior
network speed or streaming service add-ons),
promos (“Receive unlimited prepaid data to
be used for a music streaming service for only
$5 per month”), and messaging for specific
devices, locations, and events (“Upgrade
to the latest device featuring built-in VR”).
Subsequently, using audience segmentation
tools, customers can be guided to channels
that offer an engaging experience while driving
the most profitable sales outcome for the telco.
A subscriber, for example, with low-digital
propensity³, high ARPU, and high churn risk who
is living within a few miles of a store, might be a
good candidate to nudge to a device upgrade
in-store, leading to better customer experience
and potentially stronger loyalty for the operator.
Or consider a different scenario: this subscriber
uses an advanced 5G network in New York


City and is a regular user of fitness apps who
travels frequently outside the country. As a
result, her telco offers a personalized plan
recommendation with superior network access,
top fitness app subscription perks, and an
attractive international data plan.

_Case study: An Asia–Pacific operator that_
_launched a comprehensive customer value_
_management transformation powered by AI_
_(with personalization at the core) achieved a_
_more than 10 percent reduction in customer_
_churn and a 20 percent uptake in cross-sell._

Reimagine proactive service
Earlier investments in digital infrastructure
combined with predictive and prescriptive
AI capabilities enable operators to develop
a personalized service experience based on
autonomous resolution and proactive outreach.

With fully autonomous resolution, for example,
the system can predict and resolve potential
sources of customer dissatisfaction before
they are even encountered. After noticing a
customer is accruing roaming charges while
traveling abroad, the AI system automatically
applies the optimal roaming package to her
monthly bill to minimize charges. It then follows
up with a personalized bill explanation detailing
the package optimization and resulting savings
for the customer, leading to a surprising and
positive CX moment.

Operators are also exploring the redesign
of digital service journeys with the help of
AI assistants serving as digital concierges.
Generative AI technologies, including
tools such as ChatGPT, have the potential
to enhance existing bots through better
understanding of more complex customer
intents, more empathetic conversations, and
better summarization capabilities (for example,
when a bot needs to handover a customer
interaction to a human rep). A single unified
AI assistant will likely also represent a step
change in speed, accuracy, and engagement
compared to the interactive voice response
systems of today.


-----

An AI-powered service organization is a key
ingredient to releasing the full capacity of
specialized reps for high-value interactions while
improving overall customer experience—one of the
key battlegrounds for telcos around the world.

_Case study: A leading telco is expected to achieve_
_an approximately 10 percent decrease in device_
_troubleshooting calls, powered by a proactive AI_
_engine that considers the customer’s likelihood_
_of calling and issue severity to decide whether to_
_push the most effective resolution via SMS. This_
_proactive engine is also a key element of the_
_operator’s ambition to have the highest customer_
_satisfaction scores among competitors._

Build the store of the future
In retail, AI is leading a revolution in the design and
running of stores by streamlining operations and
elevating the consumer experience.

Some telcos already use virtual retail assistants
displayed on floor screens to conduct multiple
transactions with customers, including adding
balance to a prepaid account and selling prepaid
cards and TV subscriptions. A leading European
telco leverages AI tools for delivering moreaccurate device grading and trade-ins in the store.

The store of the near future includes the following
components:

— Front of house: Aisle layout and product
placement are optimized based on browsing
patterns analyzed by machine vision. Digital
signage is made relevant to individual customers
who are in-store and identified through
biometric or geofencing technology. Interactive
kiosks serve up personalized promos, service
assistance, and wait-time forecasts. Customers
are matched with reps who are given nudges
with personalized info likely to spark the
best interaction and lead to a truly seamless
customer experience.

— Back of house: Device SKUs are automatically
managed to optimize inventory and sales. Stores
stock curated assortments based on local
preferences surfaced in sales analytics. AI tools
such as computer-vision-based grading allows


for immediate price guarantees on devices that
are traded in.

— Outside: Consumers walking near the store
receive text or push notifications with a
personalized promotion and an invitation to
check the product in-store.

_Case study: An Asian telco launched a 5G virtual_
_retail assistant in unmanned pop-up stores. The_
_digital human communicates with customers_
_in a personal and friendly way with engaging_
_facial expressions and body language. She_
_supports customers across multiple transactions,_
_from buying prepaid cards to getting SIM card_
_replacements._

Deploy a self-healing, self-optimizing network
The AI-native telco will leverage technology to
optimize decision making across the network
life cycle stages, from planning and building
to running and operating. In the planning and
building stages, for example, AI can be used to
prioritize site-level capacity investments based
on granular data, such as customer-level network
experience scores.

In the running and operating phases, AI can
prioritize the dispatching of emergency crews
based on potential revenue loss or impact on
customer experience. AI can also enable a selfhealing network, which automatically fixes faults—
for example, auto-switching customers from one
carrier frequency to another because the former
was expected to become clogged. This frees up
engineering resources for higher-value-added
activities.

_Case study: A telecom operator developed an_
_AI-based customer network experience “score”_
_to improve its understanding of how customers_
_perceive their network and to inform network_
_deployment decisions. The AI engine leveraged_
_granular network-level information for every_
_line (e.g., signal strength, throughput) with an_
_ML model to create the score tailored to each_
_customer’s individual network experience and_
_expectations. The operator used the score, which_
_directly correlated with impact metrics such_
_as customer churn or network care tickets, to_


-----

_monitor network performance trending (how the_
_score fluctuated in different regions), to identify_
_opportunities to refine its buildout plan, and to_
_improve how it managed its customer base._

Improve frontline productivity
The AI-native telco also uses AI-enabled tools
to optimize workforce planning and coaching
of frontline employees across multiple teams,
including field force, customer service, and retail
associates.

For workforce planning, AI tools enhance
traditional applications by forecasting across
supply-and-demand metrics for monthly, daily, and
intraday time horizons with higher accuracy, more
granularity, and full automation. Smart scheduling
matches supply with demand, such as reps needed
in a call center during particularly busy periods, to
meet service-level targets as well as customers’
expectations.

Acting as an intelligent coaching manager, an
AI-enabled nudge engine provides personalized
celebratory and improvement opportunity nudges
to employees and their supervisors (Exhibit 1).
Coupled with advancements in generative AI, the
impact of the AI nudge engine might go even


further by, for example, simulating customer
responses under different scenarios to train reps.

_Case study: A telco operator deployed an_
_AI-enabled scheduling and coaching solution for_
_technicians servicing copper and fiber customers._
_Resulting efficiency gains included 10 to 20_
_percent capacity generation and improved_
_customer satisfaction scores._

Power intelligent internal operations
AI-powered insights will enhance decision
making across business functions, beyond the
automation of standardized or low-complexity
tasks. In finance, for example, AI can flag outlier
invoices for further inspection, while on the
accounts receivable side it can predict customers
likely to default, triggering mitigating actions.
In HR, AI can help flag employees with high
attrition or absenteeism risk and the respective
drivers while also helping identify informal
influencers who can lead change management
efforts. Generative AI solutions can help with
the development of product marketing copy, the
synthesis of customer feedback for research
purposes or even enable business users to write
simple code to quickly adjust IT applications.


Exhibit 1
**The ‘AI-native’ telco leverages AI to provide tailored coachingThe ‘AI-native’ telco leverages AI to provide tailored coaching**
**recommendations both to reps and supervisors.recommendations both to reps and supervisors.**

Illustrative call-center example

|Reps|Before each call, AI provides During call, AI assists rep with At end of week, AI generates insights/tips based of customer suggested key phrases and report with insights on rep’s profle and reminds rep of next best action (NBA) to performance and suggested best practices resolve issue coaching resources|
|---|---|


Before customer interaction During customer interaction After customer interaction

Reps

Before each call, AI provides During call, AI assists rep with At end of week, AI generates
insights/tips based of customer suggested key phrases and report with insights on rep’s
profle and reminds rep of next best action (NBA) to performance and suggested
best practices resolve issue coaching resources

Supervisors

At the start of the day, AI AI notifes supervisors of live At end of week, AI summarizes
predicts issues team may face calls that require attention with team and agent-level
and suggests resources key insights on customer performance insights and
to share in morning huddle sentiment suggested coaching resources


-----

Overall, involving AI in decision making
and execution results in higher speed and
consistency. Its benefits can be felt everywhere,
from contract management and supplier search
to onboarding and IT maintenance.

_Case study: A UK-based transportation company_
_deployed AI to identify the main drivers of_
_employee attrition and absenteeism. The_
_company then developed targeted interventions_
_for each of the drivers with an estimated 20 to 25_
_percent reduction in sick pay and attrition costs._

**Success factors of AI-native**
**transformation**
The what of envisioning being AI native is the
relatively easier part of this journey; the how of
making the possibilities reality is the tougher
challenge. Working on multiyear projects with
operators across the world, we’ve identified
critical best practices in three areas that
are the hallmarks of a successful AI-native


transformation: building AI, managing it, and
driving its adoption.

Building AI best practices
Developing transformative AI requires a
carefully calibrated approach that follows
these core guidelines:

— Build core AI capabilities in a modular
fashion and with reusability in mind, with
the potential to be deployed across
multiple contexts in the operator. A core
forecasting engine, for instance, can be
deployed both in a call center and in a retail
setting. This will drive higher ROI for AI
investments by decreasing time to deploy
and preventing duplication of work.

— Tightly integrate AI capabilities with one
another based on a model architecture
approach that interconnects different AI
models to maximize value generation and
promote reusability. For example, a digital


Web <year>

Exhibit 2 <Title>Exhibit <x> of <x>
**Digital twins create a single source of truth (‘build once, use many times’) that**

Digital twins create a single source of truth (‘build once, use many times’) that

**speeds up time to market of AI use cases.speeds up time to market of AI use cases.**

How do digital twins afect the deployment of AI?


From:
every use case curating data for its own needs


To:
one-time curating efort that is leveraged by all use cases


Traditional analytics and AI Next-generation AI use cases
Use cases that translate data into insights Applied models that drive actionable


Digital twin


Next-generation AI use cases
Applied models that drive actionable
insights/recs

Simulation and core models
Behavioral analytics and ‘what-if’
simulations

Data products
Information structured and modeled to
enable easy, reusable consumption
across needs

All data sources and platforms
Integrated, clean data and tools that
enable data governance and exploration


Traditional analytics and AI
Use cases that translate data into insights

Common data sources and platforms
Integrated, clean data and tools that
enable data governance and exploration


Data consumers tap directly into the data platform,
creating use-case-specifc data assets.
This is reactive and inefcient.


Digital twins meet multiple use cases’ needs
and are constantly evolved. This drives reuse
and ensures alignment to value creation.


-----

propensity model will be built as a core model
that becomes an input into multiple customerfacing models.

— Use digital twins as the foundation for all AI.
Digital twins—virtual representations of a
physical asset, person, or process with a data
product at its core—are the key to unlocking
reusable AI. The data in a digital twin is
intentionally structured and modeled to enable
easy, reusable consumption and governance
across needs, and to serve as the single source
of truth for all models (Exhibit 2).

— Implement machine learning operations
(MLOps) best practices to shorten the analytics
development life cycle and increase model
stability. MLOps typically involve automating
the integration and deployment of code
underlying AI capabilities.

— Rethink the tech talent strategy holistically.
Without a deep bench of engineering talent,
an AI-native ambition will remain a mirage.
Employers should consider expanding their
sourcing net to a wider range of universities
and learning environments. It’s also critical to
improve conditions that developers work under,
because developer experience is a top factor
in determining an employer’s attractiveness.⁴
Constraints on which programming languages
and cloud providers’ tools can be used, for
example, can have meaningful impact on a
developer’s decision to recruit for and stay with
an organization, as well as on the developer’s
productivity. Because tech talent needs
are multifaceted, operators should launch a
comprehensive list of initiatives across the
employee life cycle.

Managing AI best practices
Maintaining and improving AI capabilities depends
on an experimental, iterative mindset focused
squarely on product and tech innovation.


— Treat AI capabilities as true products by
assigning dedicated product managers
to oversee them. PMs act as translators
between the technical and business teams
and are mandated to own the product
continuously and develop opportunities to
improve it. They ensure that it’s never built as
a onetime solution.

— Set up AI labs for fast experimentation.
Dedicated teams of PMs and data scientists
or engineers are granted expedited approval
to experiment with new models, test for
feasibility, and validate business value before
scaling.

— Refresh the AI tech stack at least annually
to take advantage of new developments. In
recent years, there have been significant
enhancements in tooling that drastically
transformed AI workflows.

— Speed up IT and data modernization efforts
(the complexity of which often slows down
AI transformations) by leveraging reference
architectures that have been road-tested in
multiple transformations across industries.
Moreover, build the target cloud-native data
architecture following an iterative approach,
focused on enhancing the components
required for the priority AI use cases first
(for example, data streaming might be key to
unlock fraud detection use cases).

Driving AI adoption best practices
Taking a comprehensive approach focused on
both what goes into and comes out of models is
critical for fostering growing usage of AI:

— Ensure AI solutions are considered
trustworthy AI, including dimensions such
as model explainability, accountability for
the outcomes of AI models, and technical
robustness.


4 David Gibson, “New data: What developers look for in future job opportunities,” Stack Overflow, December 7, 2021.


-----

— Make change management a day one focus.
Operators need to involve end users of
AI-enabled insights through all the stages of
the model development life cycle and invest
in formal and informal capability building.
Operators will also need to take a hard look at
replacing and revamping existing processes
as well as management practices and roles to
be centered around AI.

**Next steps toward building the**
**AI-native telco**
In many industries, companies have used AI
to make their operations more efficient, drive
material enhancements in customer experience,
and ultimately used it to bring innovative
products and services to market more quickly.
Operators can learn from these industries and
invest in AI to improve their competitiveness in
the coming years of economic uncertainty and
competitive turmoil. Many operators have already
started to do so.

Organizations that talk about adopting AI but
move at a slow pace, hoping that a few innovation
projects developed at the fringes of the
organization and in silos that will come together
to create a snowball effect to holistically change
how technology informs business decision
making, are likely to fail.


Ultimately, the biggest drivers of AI adoption
will be CEO-level sponsorship and full
executive alignment throughout the AI-native
transformation. The art of the possible with
the technology has long surpassed what
companies have been able to absorb. Without
active support from the top level to proactively
address organizational inertia, communicate
an engaging change story, model new
behavior, promote capability building, and
make commitments on the required longterm technological investments, AI-native
transformation efforts will not succeed.

The journey to becoming AI native will require
operators to create a strategic vision and road
map that excites and mobilizes the organization,
build priority AI capabilities to gain momentum,
and bring everyone together to ensure
operating model and change management are
set up to drive adoption. Embracing large-scale
AI deployment across the organization will
follow.

The journey is long and requires commitment,
but operators that embrace the path to
becoming AI native are more likely to emerge as
leaders in the next horizon of transformation.


Joshan Abraham is an associate partner in McKinsey’s New York office, where Miguel Frade is a consultant and
Tomás Lajous is a senior partner. Jorge Amar is a partner in the Miami office and Yuval Atsmon is a senior partner
in the London office.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

Technology, Media & Telecommunications Practice
**How generative AI could**
**revitalize profitability for telcos**

The new technology offers the sector a real opportunity to reverse its stagnant
fortunes. But seizing it will require embracing innovation and agility to an
unprecedented degree.

_This article is a collaborative effort by Stephen Creasy, Ignacio Ferrero, Tomás Lajous, Víctor Trigo, and Benjamim_
_Vieira, representing views from McKinsey’s Technology, Media & Telecommunications Practice._

© Getty Images


-----

Amid the intense competition and cost-cutting
confronting telcos, early evidence suggests that
generative AI (gen AI) could be the catalyst to
reignite growth after a decade of stagnation. But
will it be a groundbreaking differentiator or simply
table stakes?

Already the technology is on track to become
a new norm in the industry. Most telco leaders
we surveyed[1] say they are developing gen AI
solutions that range from pilots to full-scale
deployments, and leading telcos such as AT&T,
SK Telecom, and Vodafone have made muchpublicized early gen AI commitments and
launched trials. Some telcos around the world
have started to experience significant doubledigit percentage impact from this technology.
One European telco recently increased
conversion rates for marketing campaigns by
40 percent while reducing costs by using gen AI
to personalize content. A Latin American telco
increased call center agent productivity by 25
percent and improved the quality of its customer
experience by enhancing agent skills and
knowledge with gen-AI-driven recommendations.

Most impressive is that these telcos deployed the
models in just weeks—the first went live in two
weeks, and the second in five. For an industry
with a mixed track record for capitalizing on
new technologies and legacy systems that slow
innovation, these early results and deployment
times illustrate the potentially transformative
power of gen AI.

These aren’t one-offs. Pretrained models that
can be fine-tuned in days for use cases are
readily available, enabling organizations to
bring proofs-of-concept to life with minimal
up-front investment, achieve impact out of the
gate, and scale their efforts. Our experience
working with clients indicates the potential for
telcos to achieve significant EBITDA impact
with gen AI. In some cases, estimates indicate
returns on incremental margins increasing 3 to
4 percentage points in two years, and as much
as 8 to 10 percentage points in five years, by


enhancing customer revenue through improved
customer life cycle management and decisively
reducing costs across all domains.

However, while nearly all of the 130 telcos we
surveyed are doing something with gen AI, our
survey findings suggest a palpable sense of
caution and skepticism in the industry. More
than 85 percent of the executives surveyed
are cautious to attribute more than 20 percent
revenue or cost savings impact by domain,
with the greatest enthusiasm for a radical
transformation in customer service (Exhibit 1).

This blend of optimism and restraint highlights
the critical juncture the industry faces. Seizing
the gen AI opportunity to differentiate services
and achieve sustainable growth will require
the hidebound industry to embrace innovation,
exploration, and agility at an unprecedented level
and move from decoupled AI efforts to a holistic,
AI-native telco.[2]

The chance for telcos to make this change has
never been more accessible. The industry has
struggled these last ten-plus years to achieve the
potential of “traditional” AI, given the complexity
and legacy processes involved. In addition to
the significant impact gen AI can bring to bear
with entirely new use cases and applications,
its ability to learn from vast amounts of diverse
data and interact in near-human-like ways may
be the tipping point for accelerating broader AI
programs and the building blocks that enable
them, fueling company-wide transformations.

Furthermore, the imperative for such change has
never been greater. Because gen AI democratizes
access to powerful capabilities, any telco—a
small operator or large incumbent—can reshape
customer expectations and its organizational
efficiency. In doing so, they can potentially narrow
previously unassailable competitive advantages
and overturn long-standing barriers to growth.
Those at the forefront of this movement stand to
position themselves to regain growth faster and
capture a more significant share of the nearly


1 The online survey was in the field from November 9, 2023, to December 6, 2023, and garnered responses from 130 telco operators in
North America, Latin America, Europe, Africa, Asia, and the Middle East.
2See “The AI-native telco: Radical transformation to thrive in turbulent times,” McKinsey, Feb. 27, 2023.


-----

Exhibit 1
**A large majority of telcos have already cut costs with generative AI use cases in A large majority of telcos have already cut costs with generative AI use**
**customer service and networks.cases in customer service and networks.**


Cost reduction attributed to generative AI
in diferent domains,¹ %

|Col1|Col2|2|Col4|1|Col6|1|2|Col9|Col10|
|---|---|---|---|---|---|---|---|---|---|
||6||8|2|3|1|3 15 30 50|6||
||7||||13||15|||
|||||||||6||
||||15|||||||
||32|||||||87||
||||||35|||||
||||||||30|||
||||40|||||||
||39|||||||||
||||||||50|||
||||||47|||||
||||34|||||||
||16|||||||||



Customer Network
service


0 1–5 6–20 21–50 50+ Don’t know

1 2

3 1 3 6

13 6

15

35 30

87

50
47

IT Marketing Support
and sales functions


1Gen AI CxO Survey 2023, n = 130, Q: What is the impact (% cost reduction) attributed to generative AI in the diferent domains? Percentages consider
answers only from respondents claiming to have achieved impact and to have at least some use cases in execution; fgures may not sum to 100% because
of rounding.
Source: McKinsey analysis

McKinsey & Company


**Gen AI today in the telco industry**
Gen AI represents the latest advance in AI, and
it may arguably be one of the most important.
The technology’s ability to analyze more and
different types of data such as code, images,
and text, and to create new content, enables
new levels of personalization, performance,
and customer engagement. With today’s
capabilities, many use cases are already
possible across network operations, customer
service, marketing and sales, IT, and support
functions.


$100 billion in incremental value (Exhibit 2).
That is in addition to the $140 billion to $180
billion in productivity gains that gen AI will
create in the industry above what could be
unlocked by traditional AI.

How can telco leaders use the technology
to drive AI transformations and unlock new
value? What challenges do they face? And
what will it take to succeed? This article offers
insights into these critical questions, drawing
extensively from our research, industry survey,
and firsthand experience implementing these
technologies.


-----

Exhibit 2
**Generative AI has the potential to unlock value beyond that previously Generative AI has the potential to unlock value beyond that previously**

ofered by advanced analytics and ‘traditional’ AI.

**offered by advance analytics and ‘traditional’ AI.**

Generative AI’s potential impact on global telecommunications industry,¹ $ billion


140–180

60–100 310–500

250–400 **~35–70%**

incremental

**~15–40%** economic

incremental impact
economic
impact


All worker productivity
enabled by generative AI,
including in use cases


450–680

Total AI
economic
potential


Advanced analytics,
traditional machine learning
and deep learning¹


New
generative
AI use cases


Total
use-case-driven
potential


1Updated use case estimates from “Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
Source: McKinsey Global Institute, The economic potential of generative AI, June 14, 2023

McKinsey & Company


These use cases can both enhance existing AI
capabilities (through the inclusion of new
unstructured data sources) and provide new sources
of value (through gen AI and in combination with
traditional AI solutions) to deliver significant impact
across all key domains. Customer service and
marketing and sales currently make up the largest
share of total impact (Exhibit 3).

Examples of such use cases based on early pilots
include the following:

— In customer service, where the technology can
vastly improve customer experience, increase
agent productivity, and enable fully digital
interactions. A Latin American telco is enhancing
its customer service AI chatbots to improve agent
support, a move it anticipates will reduce costs by
15 to 20 percent. The telco also is using gen AI to
summarize voice and written client interactions
in nearly a dozen use cases, with the expectation
that it can reduce associated costs by up to 80
percent.


— In marketing and sales, where gen AI enables
hyperpersonalization, deeper customer insights,
and faster content generation. A European telco
is using the technology to identify new sales
leads from customer calls, with its pilot project
achieving a more than 10 percent conversion rate.
The company can now also create personalized
messages and visual media to target individual
customer microsegments. To do this, the telco
feeds a gen AI model standard marketing
messages, customer data (including household
details, type of phone they use, and where they
live), and cognitive biases (for example, whether
the customer would be more receptive to
messaging that evokes scarcity, such as a limitedtime offer, or emphasizes authority, such as
endorsements, awards, and industry experience).

— In network operations, where gen AI can optimize
technology configurations, enhance labor
productivity, extract customer insights from
social media, and improve inventory and network
planning and management through the ability to


-----

Exhibit 3
**Gen AI is expected to enable a long list of use cases and deliver significant Gen AI is expected to enable a long list of use cases and deliver signifcant**
**value to telcos, with customer service and marketing and sales accounting value to telcos, with customer service and marketing and sales accounting**

for the largest share of total impact.

**for the largest share of total impact.**


Distribution of impact by business domain¹


Share of total impact, % Share of surveyed business
leaders focused on domain, %


Customer Marketing Support
service and sales Network IT functions


~35
~85


~15
~62


~10
~55


~5
~10


~35
~45


Example use cases

Customer-facing
chatbots, call-routing
performance, agent
copilots, bespoke
invoice creation


Content generation,
hyperpersonalization,
copilots for store
personnel, customer
sentiment analysis
and synthesis


Network inventory
mapping, network
optimization via
customer sentiment
analysis, enabling
self-healing via
customer sentiment
analysis on network
problems


Copilots for software
development,
synthetic data
generation, code
migration, IT
support chatbots


Procurement
optimization,
workplace productivity,
internal knowledge
management, content
generation, HR Q&A


1The distribution of impact by business domain is based on our experience working with telco companies to deploy gen AI and includes impacts on both capital
expenditure and EBITDA.

McKinsey & Company


mine unstructured data. One large telco is
using the technology to accelerate network
mapping by analyzing and structuring data
about network components, including
specifications and maintenance information,
from supplier contracts. This will enable the
telco to more accurately assess component
compatibility, maintenance requirements,
and more—an effort anticipated to improve
operational planning and optimize capital
productivity.

— In IT, where the technology can accelerate
software migrations and development. Gen AI
offers telcos a path to reduce their mounting
technical debt and enable capabilities
previously deferred because of time and


resource constraints. Organizations are
applying gen AI to streamline the entire
software life cycle, from documenting how
a new product, feature, or service will be
perceived by end users to generating and
scanning code for vulnerabilities before
launch. One McKinsey study found that
software developers can complete coding
tasks up to twice as fast with gen AI.[3]

— In support functions, where gen AI will
reduce the costs associated with backoffice operations and improve employee
productivity. A European telco uses the
technology in a number of ways, including:
shortening procurement analysis and
negotiation strategy insights from weeks


“Unleashing developer productivity with generative AI,” McKinsey, June 27, 2023.


-----

to a few hours, reducing recruiting costs with
automated screening and recommendations,
improving employee productivity using internal
gen AI chatbots and copilots, and automating
most internal content generation. Combined,
the company anticipates these efforts will
improve employee productivity by
30 percent.

New sources of value may also emerge from turning
internal uses cases into new products for their
customers. For example, a customer care solution
may be offered on demand to small business
customers seeking ways to improve their own call
center’s productivity and service.

**Telco leaders’ view ahead: Real change**
**and real challenges**
In the wake of their initial successes, business
leaders we surveyed say they plan to maintain
or double their budgets for gen AI in the next
year and invest in more than 50 dedicated fulltime employees to pursue their gen AI ambitions
effectively. More than half are already scaling up
use cases.

Moreover, survey findings indicate that the
technology also had a knock-on effect across all AI
initiatives. Compared to responses from McKinsey’s
2022 digital twin survey, we see a 30-percentagepoint increase in business leaders who want to
invest in and focus more on data and analytics.

However, despite the magnitude of the opportunity
and the level of interest (and need), our survey
found few who follow the kind of holistic approach
required to succeed at scale. Only about one-third
of telco leaders said they have a capability-building
plan for employees on gen AI or are investing in
change management efforts—two core building
blocks for building a culture of innovation and the
test-and-learn mindset. A similar share said gen AI
has yet to be treated as an organizational priority,
and that proponents of the technology often

4 See “Rewired to outcompete,” McKinsey Quarterly, June 20, 2023.


encounter difficulties in justifying use cases—a
clear signal that much of the push has come from
the bottom, not the top, and that more work is
needed to elevate gen AI to a CEO-led priority.
Moreover, finding appropriate talent and obtaining
quality data remain significant challenges for telcos,
although confidence about solving these rose
among surveyed leaders this year as compared
to last.

Finally, the survey findings suggest that gen AI has
already begun to influence long-standing market
dynamics. While European telecom operators
have traditionally lagged in AI and technology
transformations, survey findings indicate that
they are pulling ahead of those in North America
in their adoption of gen AI, especially in areas
such as network operations (71 percent compared
to 58 percent) and IT (67 percent compared to
55 percent). This shift may be a result of greater
maturity in managing data privacy. Small and
large operators report similar views on where to
prioritize, focusing on customer service and IT in
similar measure, suggesting the possibility of new
competitive pressures emerging for incumbents
(Exhibit 4).

**The building blocks for a successful**
**gen AI journey**
In order to achieve the above-mentioned impact,
organizations will need to move away from the
labyrinth of proofs-of-concept and scale the
technology. As with any digital or AI initiative, we
find there are no shortcuts in doing this. The same
core building blocks are necessary, namely (1) a
business-led roadmap, (2) the right talent, (3) an
operating model for scale, (4) technology built
for speed and innovation, (5) quality data that is
easily accessible and managed in a responsible
and accountable way, and (6) change management
to ensure adoption and scaling.[4] These are
fundamental pillars in effectively scaling use cases
and capturing sustainable impact from gen AI in the
journey toward an AI-native telco.


-----

Exhibit 4
**Small and large operators are focused on generative AI use cases for customer Small and large operators are focused on generative AI use cases for**
**service and IT in similar measure.customer service and IT in similar measure.**


Focus on gen AI for customer
service and IT use cases,¹ %

93


< $1 billion


$1 billion–$5 billion $5 billion–$10 billion > $10 billion


85
79


79


75


60
57
53


58 57
53 54


54
51


37


29

13
11
7 7

Network IT Marketing and sales Support functions


Customer service


1GenAI CxO Survey 2023, n = 130, Q: Focus domains in gen AI—Select the dimensions that apply depending on how you defne generative AI in your
_organization. Percentages may not sum to 100% as this question contained multiple selections._
Source: McKinsey analysis

McKinsey & Company


foundation model from scratch (what we refer to
as the “taker,” “shaper,” and “maker” approach,
respectively).

Each is suitable for different use cases and has
its own costs, requiring leaders to develop not
only a clear vision and strategy for which use
cases to pursue, but also how. One mistake we
see some telcos making is building common gen
AI solutions from scratch—a content generator or
call summarization solution for example—when
there are nearly a dozen out-of-the-box options
on the market today from gen AI start-ups or
SaaS vendors injecting gen AI capabilities into
existing solutions. Only one-third of surveyed
telco leaders say they buy products off the shelf,
suggesting that many telcos continue to embrace
a do-it-yourself model. This move is likely to
slow innovation and distract talent from more
differentiating use cases, as it has in the past with
other technologies.


However, while the same holistic approach is
required, gen AI’s unique capabilities—its ability
to surface new insights from seemingly unrelated
data, its reliance on large language models from
third-party vendors, and its transformative impact
on roles and work—present new challenges that
will require greater agility and additional oversight.
Next, we outline key differences and provide
recommendations on how telcos can best
tackle them.

Strategy: Determine when to build, buy, or finetune solutions
As with any AI initiative, leaders will need to align
on vision, value, and road map, assessing both
risks and opportunities, and communicating
guidelines for use across the organization. In
building a road map, telco leaders will face a
choice: use a commercial, off-the-shelf solution
if one exists, fine-tune existing large language
models with internal data, or build and train a new


-----

Instead, leaders should strongly consider
partnering with gen AI solution providers and
enterprise software vendors for solutions that
aren’t very complex or telco specific. This is
particularly critical in instances where any delays
in implementation will put them at a disadvantage
against competitors already leveraging these
services. The handful of solutions leaders can
concentrate on shaping or making themselves
should enable them to differentiate their offerings
or address a strategic business priority, such as
delivering the best service or network coverage,
and drive sustained economic impact. To do
this, one CIO at a large telco is bringing together
business leaders across all key domains to assess
hundreds of potential use cases and build a road
map for determining when to build, buy, or finetune models, and for prioritizing resources and
building momentum with early successes.

Talent: Upskill and expand internal expertise to
innovate with gen AI
The speed of innovation that is now possible with
gen AI puts new pressure on telcos accustomed
to outsourcing tech talent to build in-house
AI expertise. Consider the experiences of two
telcos—one that continued offshoring and
outsourcing tech talent and one that created
a dedicated AI team of ten data scientists and
engineers. In the time the first telco took to draft
requirements for outsourcing gen AI use-case
development, the second built and deployed four
gen AI solutions.

While this new technology democratizes AI by
requiring fewer highly specialized data scientists
to build the models, it requires new skills, such as
gen AI prompt engineering, which may sometimes
be a separate skill embedded within traditional
roles. It also requires significantly more data
engineers and subject matter experts who
understand what data to collect and how, and
who can oversee daily quality reviews as new
forms of data are generated and consumed by
these systems, including user queries, responses,
and feedback.


Capturing the full potential will also require
significant upskilling of existing staff—everyone
from data scientists to business leaders—on gen
AI, including the risks of uploading proprietary
data into third-party language models. Some
telcos are setting up internal certification and
university-led training programs to ensure their
teams have the right skills and capabilities to
innovate and execute with the technology. For
instance, a large telco created a badging system
to identify gen-AI-ready employees who have
completed the company’s sessions on use, risk,
and effective prompting techniques given by its
AI, legal, and risk experts. Following certification,
users participate in weekly discussion groups
to stay abreast of changes and discuss their
successes and challenges. McKinsey research
has found that such efforts improve the quality of
prompts.

Operating model: Orchestrate efforts
enterprise-wide
A significant portion of implemented gen AI
solutions can be adapted and reused in multiple
use cases. A gen AI chatbot developed to
improve agent productivity, for example, can be
repurposed with additional fine-tuning or data
to answer frequently asked questions by new
employees or provide IT support. An off-the-shelf
content generation system for drafting sales
proposals may also streamline the development
of marketing and business plans.

As a result, we’re beginning to see telcos adopt
more centralized decision making around gen
AI development. This shift includes a greater
emphasis on adopting reusable services and selfservice components, an evolution of key functions,
such as risk, FinOps, and transformation offices,
to be more focused on gen AI, and the creation
of “control towers” that can oversee all gen AI
investments and development efforts. In practice,
that can mean, for example, prioritizing the
use-case pipeline, identifying opportunities for
reusability, setting key performance indicators to
measure and track impact at the level of both use


-----

case and enterprise, and managing suppliers and
risk. A European telco’s control tower evaluates
the effects of its gen AI transformation based on
three dimensions—financial impact, user adoption,
and model performance—and aggregates the
data in dashboards that enable the company’s
top executives to keep tabs on the organization’s
progress. Similarly, a Latin American telco uses
a control tower to consolidate and standardize
supplier contracts, tracking key metrics such as
scope, duration, and renewal to compare providers
more easily, identify potential redundancies, and
reduce the manual work of digitizing content.

Technology: Create a blueprint for reusability,
innovation, and excellence
Organizations will also need a technology blueprint
that enables reusability. For instance, the blueprint
should include a framework for determining
which large language models to use and when
(commercial or open-source models, for example,
or those that support hybrid workloads). And it
should lay out how to scale a pilot—for example,
to extend a pilot that serves 100 call agents to
serve more than 10,000 agents with the same
latency and cost profile. The blueprint should
also have a framework for determining which
gen AI capabilities can be turned into ready-touse modules to be plugged into different use
cases. One large telco, for example, has begun
to identify and develop components designed
to fetch product data from a large dataset and
generate content from it that could be reused
by data science teams across domains such as
customer service, network operations, and sales
and marketing.

With new gen AI research and capabilities
being announced weekly and sometimes daily,
technology teams will also need a dedicated
gen AI innovation lab to keep abreast of industry
changes and test emerging solutions. For example,
one large telco’s chief data and analytics officer
recruited PhD graduates from universities to staff
a gen AI innovation lab and build bespoke solutions
ahead of the market to gain a competitive edge.


Once new models are deployed, telcos will need
to monitor model outputs daily to ensure quality
and accuracy do not waver as models learn and
adapt their responses based on user queries
and feedback. Large language model operations
(LLMOps) is an emerging practice that aims to
streamline the daily management and monitoring
of gen AI models. A key component of LLMOps is a
dedicated operations team to oversee all deployed
gen AI models, continuously monitoring for issues
and rapidly adapting solutions when needed, just
as a network operations team might do for network
performance. Organizations can start small
now and build capability in this area as the field
of LLMOps develops. For example, a European
telco started by assigning three data scientists to
monitor its handful of deployed models and plans
to expand the team as more models are deployed.

Data: Capture everything, especially
unstructured data, and ensure responsible use
One of gen AI’s superpowers is its ability to
uncover connections in seemingly unrelated
datasets, which has implications for how
organizations choose to collect and measure data,
and how they manage it to ensure responsible use.

_Data collection: Telcos will need to think more_
broadly about data collection, mapping more
data, setting up pipelines for unstructured data,
and creating synthetic data to evaluate outputs.
A US telco, for instance, has reached beyond
its traditional datasets in its work to develop a
customer service agent copilot that will reduce
average resolution times by 40 percent across
more than one million annual chats. As part of
their work, the company’s data scientists gather
institutional knowledge from agent emails and
interactions to enable the chatbot to learn from
real situations and challenges, and offer detailed
descriptions of how to resolve specific issues.
The team also creates synthetic data using a
large language model to create sample customer
questions and answers, with agents reviewing the
outputs for accuracy.


-----

_Responsible use: These types of more_
sophisticated data strategies and tactics come
with new regulatory, intellectual property, and
data privacy concerns. Risks abound in this
new era, particularly with customer insights,
recommendations, and network optimizations
being analyzed and generated by third-party
large language models and open-source
environments. To address the novel risks, telcos
need to expand their data governance programs
to address unstructured data. For example, one
multinational telco hardware provider created
a robust data access process to validate what
data can be used in gen AI use cases. Data
owners along with legal and security experts
work together to validate each use case based
on several criteria including the criticality of the
data to the business (data that is deemed of high
importance cannot be input into commercial large
language models); the end users (some users
cannot access certain data assets); and the risks
if the gen AI solution gives an incorrect answer.
The team manages the process in an agile manner
using a simple Microsoft Power App to manage
and automate the workflow across teams, and
conducts monthly forums to review the process
and develop improvements. The organization has
reviewed more than 200 use cases, rejecting a
number due to intellectual property and other
risks, to ensure responsible use for the company.

Change management: Ensure adoption and
scaling are CEO-led
Every role, including everyone from network
technicians to HR professionals, will be impacted
by gen AI, making vital the need for leaders to
begin preparing their employees now to capture
the full value of this transformative technology.
With many employees already using the
technology in their personal lives, organizations
will need to consider how to help them learn to
apply the technology in a professional context,
upskilling and reskilling staff at scale. Such
work can be made easier using gen AI, for
example to develop and deliver customized and
adaptive training programs, and even to onboard
employees.


For example, another European telco saw
firsthand the importance of change management
and upskilling when it created a gen-AI-driven
knowledge “expert” that helped agents get
answers to customer questions more quickly.
The initial pilot, which didn’t include any process
changes or employee education, realized just
a 5 percent improvement in productivity. As
the organization prepared to scale the solution,
leaders dedicated 90 percent of the budget
to agent training and change management
processes, which facilitated the adoption of the
solution and resulted in more than 30 percent
productivity improvement. The telco also used
gen AI to create upskilling programs and provide
agents with personalized recommendations for
improvement once the solution was rolled out.

Even though so many companies have already
achieved real cost savings and revenue
improvements with gen AI, these are still the early
days of the technology. In the next five years,
emerging capabilities—including significant
improvements in natural language understanding,
advances in human-like reasoning across
multiple topics, and availability of real-time
solutions with increased accuracy and fewer
hallucinations—should unlock even more exciting
opportunities beyond the basic improvements
seen today.

Combined, these gen AI capabilities will enable
telcos to redefine industry standards and set
themselves apart in the market. For example,
network operations could be enhanced and
quality standards radically recast with AI copilots
that evaluate images from technicians, provide
accurate recommendations for remedies, and
automatically initiate interventions or work
orders. In sales, cognitive copilots could conduct
sentiment analysis on customer calls in real
time and guide sales representatives on how
best to respond, profoundly altering sales
strategies, customer engagement, and overall
sales outcomes. Customer service channels


-----

using cognitive chatbots could seamlessly
answer complex queries in real time while taking
into account privacy and fairness concerns,
thereby revolutionizing efficiency while offering
customers a human-like experience. Across the
enterprise, greater efficiency and productivity
could emerge, as domain-specific solutions
endowed with an organization’s institutional
knowledge power an unprecedented wave of
automation and AI-driven decision making.

The sudden rise of gen AI has brought the
dream of the AI-native telco significantly


closer to becoming a reality. With it comes the
opportunity for telcos to reverse their recent
stagnant fortunes and usher in a new era of
growth and innovation. The journey will not be
easy, however. To answer the call of gen AI, telcos
will need to quickly adopt a culture of innovation
and experimentation enabled by the core
building blocks shared in this article, one they
have previously struggled to build and maintain.
With the technology moving so rapidly, those
operators that embrace it now are likeliest to
create a significant lead that will be difficult for
others to follow.


Stephen Creasy is a partner in McKinsey’s Copenhagen office, Ignacio Ferrero is a partner in the Miami office,
Tomás Lajous is a senior partner in the New York City office, Víctor Trigo is an associate partner in the Madrid
office, and Benjamim Vieira is a senior partner in the Lisbon office.

The authors wish to thank Joshan Cherian Abraham, Eric Buesing, Michael Chui, Guilherme Cruz, Sebastian Cubela,
Andrea Fariña, Roger Roberts, and Kayvaun Rowshankish for their contributions to this article.

Copyright © 2024 McKinsey & Company. All rights reserved.


-----

Technology, Media & Telecommunications Practice
**Generative AI use cases:**
**A guide to developing the**
**telco of the future**

To maximize the opportunity from generative AI, telcos and other TMT players
should pursue use cases that have the greatest chance of success and can
position the organization for growth and innovation.

_This article is a collaborative effort by Ignacio Ferrero, Víctor García de la Torre, Tomás Lajous, Víctor Trigo, and_
_Benjamim Viera, representing views from McKinsey’s Technology, Media & Telecommunications Practice._


© Getty Images


-----

Generative AI’s impact on telecom operators is likely to be between $60 billion and $100 billion,
according to McKinsey estimates. To capture this new value, organizations will benefit from
developing a strategy and implementation road map of use cases across the entire organization.
This road map should encompass key areas such as marketing and sales, customer experience (CX),
customer service, IT, networks, and support functions. At successful companies, implementation
involves a careful choice of priorities and critical decisions based on expected impact and feasibility.
Successful companies also tend to favor quick wins to start building capabilities, including the main
building blocks: governance, data, talent, and technologies. Strengthening this core enables the
move from proofs of concept to use cases at scale.

In the following exhibits, we examine the use cases that hold the most promise for telcos and apply
to the broader technology, media, and telecommunications (TMT) sector. Individually, these use
cases deliver incremental improvements on existing processes. Taken together, they add up to a
radical reimagination of the telco of the future.

This is only the beginning. As gen AI continues to develop and mature, it will unleash even more
exciting approaches to fostering creativity, innovation, and growth.

**Telco vision: Gen AI will drive the next wave of productivity and**
**innovation**
The technology is poised to unlock an array of new efficiencies, improving the way telecom
operators attract customers, provide service, maintain networks, and develop software and
systems—potentially reigniting growth after a long period of stagnation (Exhibit 1).


-----

Exhibit 1
How generative AI will accelerate the telco of thefuture.How generative AI will accelerate the telco of the future.




​B2B marketing and sales ​B2C marketing and sales IT

productivity

5–15% in marketing 3–5% sales

20–40% [productivity2]

functions1

Pre/post-sales activities are AI-​ Enhanced CX across the customer life
driven, with enhanced lead generation cycle, driven by hyperpersonalization, Higher productivity as AI-assisted tools
via market research and hyper- better digital interactions, and ad hoc accelerate tech delivery timelines with
personalized product oferings product ofering and communications code development copilots (25–30%
and proposals tailored to (eg, AI-driven content, customized reduction of time to market), automated
specifc companies. bundles, and conteitual promotions/ documentation, automatic testing, and

interactions), leading to conversion rate proactive recommendations.

Negotiation is customized per ​ uplift of ~10–15%.
customer and copiloted with gen AI to New frontiers of data, achieved by
enhance customer eiperience (CX) enhanced value eitraction and insight
and boost sales. creation from unstructured data (eg,

​Customer service PDFs, social media, audio, images).

utreach communications (eg, emails ​
and landing pages) are automatically
created, leading to cost reduction Up to reduction in humanof ~5–10%. 50% serviced contacts Network

Support functions

Reimagined CX journeys with chatbots able to deliver real-time personalized 15–20% capital expenditures
responses to customer queries,

20–30% [automated ] leading to a 30–5% reduction in
HR tasks

function costs.

Decision making across all stages of

Agents leverage AI-developed scripts network life cycle, from plan and build

Improved productivity across the value and receive instant feedback, to run and operate, optimized by
chain from speed, automation, and accessing relevant customer data leveraging AI.
insight creation via tools like HR for tailored and real-time
knowledge bots and enhanced information delivery. Enhanced maintenance with self
healing processes to automatically fi

screening of profles.

technical issues and boost productivity
of engineering resources (eg, a 25–
30% reduction in size of incident
management team).

1Including efcient and efective content creation, product discovery and search personalization, and SE optimization, as well as reduction in spending on eiternal
channels and agencies.

2Direct impact of AI on the productivity of software engineering of current annual spending on the function.

Source: Eipert input; The economic potential of generative AI, McKinsey Global Institute, 2023

McKinsey & Company

**Telecom and other TMT players should develop a comprehensive road**
**map of gen AI use cases across the organization**
Early movers are building proofs of concept with minimal up-front investment and realizing
significant returns (Exhibit 2). In some cases, estimates indicate returns on incremental margins
increasing three to four percentage points in two years and as much as eight to ten percentage
points in five years.


-----

Exhibit 2


**The first step to becoming an AI-native telco player: Identify use cases across the**
**organization.**


-----

**Gen AI can boost sales by automatically producing personalized content**
Gen-AI-accelerated content creation enables hyperpersonalization across campaigns, enhancing the quality
of customer interactions on all channels while improving the effectiveness of top-line efforts and optimizing
marketing budget allocation (Exhibit 3).

Exhibit 3
**B2C marketing and sales use cases: Bespoke offerings, smarter campaigns.**

|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|---|---|---|---|
|Hyper- personalized content|​Boost marketing efforts by tailoring communications to specific customer profiles. Gen AI generates text/audio/ images and adapts them to specific users (eg, creating more engaging content by adapting language and tone to match user preferences)|Create hyperpersonalized text/email communications (eg, a personalized text to young female with 2 mobile plan subscriptions who lives in the southern US and whose favorite app is TikTok, with key features based on this data)|15–30% increase in conversion rate (CR)|
|Intelligent content generation for marketing campaigns|Automatically create marketing content, including images and videos, for ads, social media posts, landing pages, email campaigns, and other marketing channels|Generate engaging marketing and social media content for new pricing promotions for mobile plans (eg, personalized images for new handset marketing campaign)|10% increase in CR|
|Enhanced global marketing reach through dynamic copy translation|Gen AI’s ability to understand context enables automatic translation of marketing copy, supported by quality assurance measures (eg, checking for grammatical errors, clarity, and overall coherence)|Real-time translation of chatbot answers for prepaid international customers and translation of marketing copy in different regions|5–15% increase in productivity|
|Tailored product offerings|Gen AI enhances customer experience with custom product offers, generating new bundles for each customer, based on segment features and historical data|Tailored offer to adults aged 20–30 who consume more gigabytes and fewer call minutes|5% increase in CR|
|Customer feedback analysis|Gen AI enables the extraction of information from unstructured customer feedback sources (eg, call center, web forms, chatbots, etc) to generate insights and identify issues for further analysis and resolution|On the launch of a new technology device, monitor consumer sentiment by automatically analyzing social media posts|10-point increase in customer satisfaction|


-----

|Exhibit 3 continued|Col2|Col3|Col4|
|---|---|---|---|
|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|Unlocking best practices of sales champions|Identify specific actions employed by top-performing agents and share them to enhance performance|Upskill sales force (eg, outbound call center sales teams, SME² team) by extracting best practices for communication techniques, relationship- building strategies, and persuasive selling approaches|10–15% increase in CR|
|Smart campaign investment optimization|Optimize the investment mix of campaigns deployed, optimizing conversion rates|Improve the budget allocation of lead campaigns across channels|2–3% increase in leads, calls, and sales|


1 Not considering cross-effects and interdependencies between different use cases.
2Small and medium-size enterprise.

**Gen AI can optimize call center efficiencies while enhancing the customer**
**experience**
Gen AI helps customers get answers and solutions faster. A major reason is that it supports the work of agents by
simplifying their access to relevant data, personalized recommendations, and best practices (Exhibit 4). With this
support, agent performance is expected to improve rapidly across the board.

Exhibit 4
**Customer service use cases: Smarter agents, happier clients.**

|Prioritized use case|Future state: What this could look like|TMT application|Potential impact¹|
|---|---|---|---|
|Proactive root-cause analysis|Leverage advanced analytics to examine call center conversations, identifying recurring issues and root causes to enhance call center operations and improve overall customer experience|Proactively address major reasons for customer care calls, enhancing customer satisfaction (eg, billing issues, roaming coverage)|30–50% call reduction|
|Enhanced agent coaching|Empower agents with improvement opportunities through personalized nudges and hyperpersonalized training content (eg, identifying mistakes in customer interactions) to highlight best practices to enhance the next best action for agent training|Enhance call center agent training to improve performance in future interactions by identifying areas of improvement and provide personalized content (eg, tips to improve script about launch of a new post-paid bundle campaign)|10–15% cost savings in onboarding and training|


-----

Exhibit 4 continued

|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|---|---|---|---|
|After-call log- in automation|Gen AI enables the extraction of information from unstructured data (eg, call recordings, post-call notes), automatically extrapolating pertinent information and updating the CRM system|After a customer-agent interaction, such as the canceling of a subscription plan, an automatic ticket is created, along with the identification of root causes and other key attributes|20–30% increase in typified calls|
|Hyper- personalized customer service chatbots|Gen AI enables a 1:1 personalized chatbot experience that proposes an ad hoc solution path, leveraging the customer’s existing information and history|Increase the capabilities of the digital sales channel by incorporating upsell, cross-sell, and personalized attention in chatbots, trained with each customer’s previous interactions (eg, can be connected to CRM to add additional features)|10–20% call reduction|
|Invoice- focused chatbot|Gen AI is capable of extracting information from invoice documents and providing comprehensive and clear explanations of invoices|Explain increases in price to customers by comparing invoices from last 6 months (eg, customer contracted a plan with more gigabytes)|5–10% call reduction|
|Customer interaction database|Gen-AI-driven database that stores relevant information related to each customer- agent interaction, including summaries of intent, outcome, and resolution path; this can be achieved by leveraging previously untapped data sources, such as call transcripts or audio recordings|Streamline and improve quality of call center agents by facilitating performance evaluations and supporting decision making enabled by easy retrieval and review of call details (eg, classification of customer intent)|50–60% reduction in after-call work (ACW)|



1 Not considering cross-effects and interdependencies between different use cases.


-----

**Gen AI can boost network operations and reduce call center costs by reducing the**
**number of customers experiencing technical issues**
Gen AI’s ability to act on unstructured data helps teams optimize network operations and reduce downtimes
(Exhibit 5). And stronger processes allow teams to act faster and more effectively.

Exhibit 5
**Network use cases: Fewer outages, faster answers, swifter repairs.**

|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|---|---|---|---|
|Circuit inventory|Use data from multiple sources to consolidate information and create a unified vision of end- to-end coverage of operator networks and add additional specifications (eg, network coverage, service availability, and technical qualifications)|Create a unified vision of the network’s end-to-end coverage (eg, wirelines), leveraging gen AI along with existing unstructured data (eg, supplier contracts)|2.5–5% reduction in future contract costs|
|Chatbot for network data|Deploy an internal chatbot with network information from multiple sources|Assist in organizing the network team’s information and responding to other department queries|10–15% increase in productivity of network employees|
|Root-cause identification|Diagnostics and resolution assistant that acts as experts’ copilot to enhance capabilities and efficiency in identifying root causes and proposing solutions|In an outage, the system automatically runs a diagnostic analysis to identify root cause, helping network admins resolve the issue quickly|6–12% reduction in customer care tickets|
|Self-healing network|Provide assistance to agents or bots in programming routers for efficient and proactive network management and troubleshooting|In case of a Wi-Fi issue, automatically generate a repair process (eg, reboot network) adapted to contextual data, before dispatching technicians|20–30% increase in productivity|
|Mobile care|Enable proactive actions to prevent and enhance complaint management|Refine the model to identify incidents accurately and integrate with customer care operations|30–35% decrease in call center calls related to massive incidents|


-----

|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|---|---|---|---|
|Optimization of own vs third- party network usage|Optimize the use of own network by flagging changes or errors in external network parametrization|Add a mechanism for automatic detection of deviations of NRA, RaaS, and VAS traffic between the Comp. C billing platform and the network data|0.3–0.8% reduction in third-party network consumption|


Prioritized Future state: What this Telco/TMT Potential
use case could look like application impact¹

Add a mechanism for

Optimization Optimize the use of own automatic detection of 0.3–0.8%
of own vs third- network by flagging changes deviations of NRA, RaaS, reduction in

third-party

party network or errors in external network and VAS traffic between the

network

usage parametrization Comp. C billing platform and consumption

the network data


1 Not considering cross-effects and interdependencies between different use cases.

**Gen AI can be deployed to optimize IT operations and streamline the entire software**
**life cycle**
Already, organizations are using gen AI to accelerate software development, testing, and migration—which could
free up time and resources for enabling capabilities previously deferred (Exhibit 6).

Exhibit 6
**IT use cases: More productive developers, smarter systems.**

|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|---|---|---|---|
|Code debugging and optimization|Accelerate tech delivery timelines with automated code development tools enabling low/no-code development (eg, natural-language coding)|Provide support on code review and pull-request completion|40–55% increase in productivity|
|Software development copilot|Accelerate tech delivery timelines with automated code development via gen-AI-driven tools (eg, GitHub Copilot); Copilot offers intelligent code suggestions and autocompletion capabilities and debugging support and documentation|Leverage GitHub Copilot to develop/optimize new code|25–30% increase in productivity|


-----

|Prioritized use case|Future state: What this could look like|Telco/TMT application|Potential impact¹|
|---|---|---|---|
|Code migration for legacy languages|Interpret, translate, and generate code (eg, migration from legacy systems at scale, automating test development, generating documentation, and performing linting)|Translate COBOL code to Java|20–30% increase in productivity|
|Query assistant|Assist in efficiently and accurately querying (eg, SQL) databases and retrieving specific data, thereby improving data retrieval and analysis processes|Customer value management (CVM) team can independently extract information (eg, new handset sales) to monitor campaign performance|20–30% increase in productivity|


1 Not considering cross-effects and interdependencies between different use cases.

Ignacio Ferrero is a partner in McKinsey’s Miami office, Víctor García de la Torre and Víctor Trigo are associate partners in the
Madrid office, Tomás Lajous is a senior partner in the New York office, and Benjamim Vieira is a senior partner in the Lisbon office.

The authors wish to thank José David Vázquez Zelaya for his contributions to this article.

Copyright © 2024 McKinsey & Company. All rights reserved.


-----

Technology, Media & Telecommunications Practice
##### Tech talent in transition: Seven technology trends reshaping telcos

With technologies like edge computing, AI, and xRAN transforming telecom,
leaders are reassessing how to capture value. Rethinking talent is a key piece
of the puzzle.

_by Tomás Lajous, Stephanie Madner, Carlo Palermo, and Rens van den Broek_


-----

The telecom industry is evolving quickly, as
businesses and consumers seek out gamechanging use cases—from autonomous vehicles
to robotic surgery to an unfathomable range
of seamless digital interactions—that operate
on the back of telcos’ substantial 5G
infrastructure investments.

Telco leaders are broadly aware of the magnitude
of transformation that the moment demands, and
many are creating elaborate plans to overhaul
everything from business models to operations to
customer experience. Ongoing excitement about
the potential of AI, driven by advances in generative
AI, is pushing the industry to rethink the scope of its
transformation plans. However, many telco leaders
are struggling to manage the talent implications of
these shifts, including determining what talent they
need and how to beat out the competition to get it.

The industry has certainly not lacked for
engineering PhDs or other markers of technical
acumen over the years. But the tech talent market
and telcos’ position within it have changed
dramatically since the generation that is now on the
brink of retirement embarked on their careers.

Moreover, not all tech talent is created equal. As
telcos evolve to deliver on the opportunities that
AI, augmented and virtual reality, and other
emerging technologies unlock, they will need to be
highly strategic about identifying and attracting
talent with the expertise and abilities that each
technology demands.

To frame the path ahead, we outline seven broad
tech trends that are reshaping the telco industry,
along with the talent implications of these trends—
including the specific skill sets and capabilities
required, as well as those that will likely be
phased out. These current tech trends create an
urgency for telcos to act now and identify critical
talent pools to develop.


We then offer an approach to guide telcos through
the complex process of fulfilling their immediate
and long-term talent needs. While this approach is
rooted in the present trend landscape, it is
designed with adaptability in mind and as such will
be relevant and applicable to future tech trends
that may rise in prominence.

The terrain here is not friendly. Long gone are the
days when telcos were the employer of choice for
technical talent. Over the next decade, demand for
certain tech roles is expected to further increase 20
to 30 percent across US industries—potentially
outpacing the supply of recent STEM graduates,
which grew just 5 to 10 percent annually from 2015
to 2019. For some roles, telcos’ demand is expected
to outstrip that of other industries: by 2031, for
example, telcos’ demand for electrical engineers is
expected to grow 24.4 percent, compared with
5.9 percent in other sectors (Exhibit 1).

Telco operators with ambitious goals regarding
diversity, equity, and inclusion should be particularly
intentional about developing sustainable, long-term
talent pipelines. McKinsey research shows that
diverse organizations increasingly outperform their
nondiverse peers. Telcos’ current tech talent pools
tend to be less diverse than their overall talent
pools; if operators’ current talent acquisition and
development patterns continue, they stand to
become even less diverse overall as their tech
talent pipeline grows.

**Seven tech trends shaping telcos**

As digital transformation continues to accelerate,
we are on the cusp of further seismic changes to
how we work, live, travel, and interact. The seven
trends described below are poised to redefine


-----

1. Ever-expanding connectivity
Fifth-generation (5G) telecommunications
infrastructure is dramatically expanding and
improving connectivity, and sixth-generation (6G)
infrastructure is poised to amplify this trend.

Near-limitless connectivity will pave the way for
new services like remote patient monitoring and
next-generation customer experiences like virtual
dressing rooms and conferences that take place
entirely in the metaverse.

Demand for connectivity is expected to increase
further as customers seek out these innovative
solutions and as the number of connected devices
grows to a projected 51.9 billion by 2025,[1] up from
43 billion in 2020. Remote work is also fueling


customers’ expectations of telcos—and the role
that telcos can play in individuals’ lives and the
success of organizations.

Each technology will require telcos to grow and
stretch in new ways, compelling telco leaders to
determine early on where to place bets and to
continually refine their priorities as the landscape
shifts and technology evolves further. As telcos hire
the talent needed to embrace the seven tech
trends, they will have less need for skills that can
now be automated or are specific to outmoded
legacy infrastructure.

At every stage, having the right talent in place
will distinguish the leaders from their less
successful peers.

Exhibit 1


-----

demand, with 51 percent of Americans working from
home at least one day a week.[2] Moreover, 5G and
6G are estimated to expand connectivity to up to
80 percent of the global population by 2030.[3]

To meet this demand, telcos will need to
exponentially increase network capacity, improve
data throughput and spectrum, and reduce latency
and energy consumption. In addition to extending
coverage to individuals, telcos may have an
opportunity to raise B2B revenues by developing
premium connectivity solutions for specific
use cases.

This will require talent with skills in network and
_spectrum design to work on strategy and_
architecture; network engineering, to design
architecture and develop applications; network
_innovation, to develop emerging RAN (radio access_
network), network functions virtualization,
Kubernetes, etcetera; network maintenance
_monitoring, to handle emergencies and to fix_
breaks; and IoT, to develop applications, platforms,
and APIs.

_Engineering and operational capabilities specific_
to legacy technologies, such as digital subscriber
line (DSL), 2G and 3G cellular networks, and
traditional cable TV infrastructure, will likely no
longer be needed. The talent base, therefore, will
need to adapt.

2. Edge computing
As computing workloads are distributed across
remote data centers located closer to end users,
latency will drop, bandwidth will increase, and
organizations will gain more sovereignty over their
data. Edge computing allows for real-time data
processing, which will unlock use cases across
industries—from remote healthcare treatment to
remote management of mining operations to
sustainability solutions like smart grids that
optimize energy consumption.


A recent McKinsey survey of 75 telco executives
across North America and Western Europe
detected a great deal of interest in a variety of edge
computing use cases (Exhibit 2). The survey results
showed that a majority of telcos are engaging with
edge computing on some level, with a quarter
already deploying it or actively planning to scale it.
More than half of the executives (55 percent)
surveyed said their primary goal is improving
network efficiency and performance, while others
cited enabling new use cases for businesses
(21 percent) or for consumers (18 percent).

As telcos embrace edge computing, they will face
rising costs from energy consumption, network
maintenance, and investments associated with
reconfiguring network backhaul and backbone.

The move toward edge computing requires telco
talent with skills in network and system design to
work on data strategy and architecture; network
_engineering, to enable the installation and_
integration of devices, software, and systems;
_network innovation, to improve the performance of_
systems; network maintenance, to fix breaks and
handle emergencies; database management, to
manage data storage, distribution, and analysis;
and security, to minimize fraud, monitor risk, and
handle compliance.

The rise of cloud-based solutions, automation, and
managed services will reduce demand for on-site IT
_setup and maintenance roles._

3. Next-generation transportation
The first two tech trends, expanded connectivity
and edge computing, lay the groundwork for a third:
next-generation transportation. The shift toward
autonomous, connected, electric, and smart
technologies has vast implications for air and land


2 “Americans are embracing flexible work—and they want more of it,” McKinsey, June 23, 2022.


-----

transportation, with the potential to make human
travel and the transport of goods far more efficient
and environmentally sustainable.

The transportation industry will increasingly
prioritize electric, hydrogen-based, and hybrid
propulsion as new modes for ground and air
mobility. An expected rise in data traffic and

Exhibit 2


autonomous-landing applications may allow
businesses to expand their markets, reaching
new customer segments in previously
unserviceable locations.

As transportation evolves, telcos will need to
increase bandwidth for mobility, particularly in
remote areas, and provide flawless emergency


-----

backup coverage. They will also have an
opportunity to combine core connectivity with
vehicular technologies and real-time mobility data
to offer solutions like hands-free driving,
infotainment, networks of smart electric vehicle
chargers, and “vehicle to everything” (V2X)
technology—which allows vehicles to connect with
their surroundings, including other vehicles and
human drivers.

Telcos will need talent with skills in network design
to develop algorithms for vehicle connectivity;
_network engineering, innovation, and maintenance,_
to enable vehicle-to-infrastructure connectivity;
_automation, to leverage machine learning and AI for_
infotainment; IoT architecture, to enable voice
recognition and gesture control; UX design, to
enhance the user experience; and data science, for
collecting and processing data.

4. xRAN
New approaches to RAN can bring flexibility to
telcos’ relationship with OEMs and even reduce
physical-asset requirements like towers, antennas,
and cabling, thereby cutting capital and operational
spending, accelerating the deployment of new
network services, and spurring competition
among vendors.

Falling under the umbrella of “xRAN,” these new
approaches include open RAN (ORAN), which will
allow for seamless interoperability among hardware
and software from disparate vendors when it
reaches maturity; centralized RAN (CRAN), which
allows multiple mobile sites to share equipment;
and virtualized RAN (VRAN), which supports
scalability and network agility by decoupling
network hardware from software.

xRAN has the potential to improve telcos’ total cost
of ownership by allowing them to choose different
suppliers for different needs—a dynamic that may


encourage new vendors to enter the market and
lead to more competitive pricing. This flexibility may
lower the risk telcos face when they adopt new
hardware or software solutions. And the availability
of intelligent, virtualized, and interoperable
functions will allow organizations to assemble
tailored solutions that increase their capacity.

Our survey of telco executives indicates a strong
interest in ORAN in particular, with 76 percent of
executives at incumbent telcos and 88 percent of
executives at new entrants planning to invest in the
new approach. Overall, 60 percent of executives
indicated that they plan to use ORAN for at least
20 to 30 percent of new network build-outs.

To accomplish this, telcos will need talent skilled
in agile working to enhance engineering practices
and innovation deployment; data engineering,
to develop architecture; cloud, to develop and
test solutions that xRAN enables; product
_management, to enable the evolution of xRAN;_
and DevOps, to build solutions and accelerate the
transition to xRAN.

There will be less need for proprietary hardware
_knowledge, closed system integration skills, and_
_manual operational capabilities specific to legacy_
_RAN systems._

5. Trust architecture and digital identity
As organizations build and scale digitally enabled
products and services that hinge on collecting vast
troves of customer data, trust and privacy will
become even more essential. Zero trust
architecture, digital identity, and privacy
engineering will become more prevalent as
companies seek to gain a competitive edge by
establishing stakeholders’ trust.

To meet rising consumer expectations around
digital trust, IT security, and data visibility, telcos


-----

history, and other considerations. By analyzing
customer behavior trends, generative AI can
enhance product development and accelerate
innovation; it might suggest new features for a
mobile app or new plans targeting specific
customer segments. By using generative AI to
simulate sophisticated cyberattacks, operators
can identify vulnerabilities and enhance
network resilience.

To maximize the AI opportunity, telcos will need
talent with skills in interface design to create
excellent user experiences; natural language
_processing engineering, for AI speech recognition;_
_data engineering, to work on data architecture,_
software, and big data; data science, to create
mathematical machine learning models; and
_security, to prevent and manage cyberattacks._

As infrastructure is increasingly managed through
software, AI will supplant the need for routine
_manual troubleshooting._

7. Quantum technology
Our survey reflects broad consensus among telco
executives regarding the impact of quantum
technology, with 52 percent saying they believe
that quantum will be a differentiating advantage
for telcos in the next five years (and an additional
32 percent saying they somewhat agree with
this assessment).

Executives see the highest strategic value in
developing quantum key distribution (QKD)
networks, which allow for the secure exchange of
cryptographic keys. Roughly half of executives are
already engaging with quantum technology to
protect customer data or improve procedures for
authenticating users’ IoT devices (55 percent),
protect telco infrastructure through encryption
(53 percent), or encrypt traffic within the network
(48 percent).


should consider investing in cybersecurity
solutions. Those that do so will position themselves
to introduce new offerings by building digital
identity services on next-generation networks
and technologies.

To realize this potential, telcos will need talent with
skills in digital identity development to provide
solutions and trusted technologies; cybersecurity
_solution architecture and engineering, to ensure_
assessment and secure access to networks and
applications; automation, to create digital identity
solutions and tools; privacy engineering, to handle
risk and compliance; network engineering, to
develop apps and architecture; network
_maintenance, to monitor and manage emergencies;_
and DevOps, to automate configuration, continuous
delivery, and infrastructure.

_Manual production and review of compliance_
_documentation will be phased out._

6. Artificial intelligence
Advances in AI—and in generative AI, in
particular—are unlocking opportunities for
organizations at every point along the value chain.
Telcos can use AI to optimize networks (by
managing resources based on real-time traffic and
data analysis); proactively address maintenance
issues (by analyzing patterns and anomalies to
identify problems before they occur); and minimize
churn (by analyzing customers’ behavior to identify
those most likely to leave). By coupling AI-powered
cameras and sensors with AI-enabled network
maintenance automation, telcos can substantially
reduce the costs associated with network
infrastructure management.

Generative AI can transform customer experience
by supplying customers with highly personalized
content, offers, and proactive service-related
outreach based on usage patterns, purchase


-----

At the same time, quantum computing will put
conventional encryption methods at risk by opening
new attack vectors. Organizations are already
growing concerned about “harvest now, decrypt
later” attacks, in which bad actors steal encrypted
data in hopes of using quantum computers to
decrypt it in the future. By harnessing quantum
technology, telcos can equip themselves with tools
to combat these sophisticated threats; QKD, for
example, allows communicating parties to be
alerted any time an intruder attempts to eavesdrop
on an encrypted exchange.

Advances in quantum technology also have the
potential to exponentially increase computational
performance and the speed of communication.
But despite telco leaders’ enthusiasm, very
few organizations are actively deploying quantum
at scale.

To move beyond internal discussions and test-andlearn pilots, telcos will need talent with expertise in
_quantum technology (such as quantum algorithms,_
_computer architectures, superconducting circuits,_
_and machine learning); high-performance_
_computing, to engage with the ecosystem on pilots_
in areas like QKD; software and hardware security
_and crypto-agility, to prevent cyberattacks and_
manage cryptography transformations as threat
levels and standards evolve; network engineering,
to design hybrid classical/quantum networks,
codevelop and pilot critical equipment for optical
communications, and explore the potential of
satellite and fiber for quantum communications; and
_product management, to monetize quantum_
networks and security.

_Engineering and operational capabilities specific to_
_traditional network optimization methods, using_
classical computation, will become less relevant.


**Designing a telco talent road map**

To get ahead of these seven tech trends, telcos will
need to develop long-term strategies for
cultivating, attracting, and retaining the right talent,
with the right skills. Leading organizations will
prioritize diversity at every stage, from strategy
design through implementation. They will also
engage business leaders from the outset, ensuring
that they help shape the tech talent strategy—and
that they own it.

Phase one: Identify the talent implications of
business strategy
When it comes to incorporating new technologies,
one of the most common challenges telcos face is
targeting investments directly to those capabilities
that fully align with their broader business goals. In
the context of a telco’s business strategy and
competitive landscape, some of the seven tech
trends outlined above may be more beneficial or
immediately relevant than others.

Telcos can start the process of getting the right
tech talent in place by defining their vision for
business success over the next three to five years.
They can then conduct thorough business impact
assessments of the seven tech trends to evaluate
how each trend might fuel ambitions like expanding
market share, enhancing customer experience, or
increasing operational efficiency. Telcos can then
work backward to pinpoint the skills and
capabilities required to lean into the technologies
most pivotal to meeting business objectives. Some
examples of the different approaches a telco might
take depending on its primary goal include:

— A telco aiming to boost its B2B leadership

might opt to invest heavily in edge computing
and quantum technology. These tech trends


-----

assessments of how demand for each role will
change. For operators that are serious about
diversifying the workforce, it will be important to
understand how demographic variations play out
across the talent pool.

At this stage, telcos can also look at broader shifts
in supply and demand for different types of roles
across the economy. In addition to examining how
their hiring rates must evolve, it will be important to
consider the sheer numbers involved.

Here are some questions telcos might consider
when determining which talent pools to prioritize:

— What is the business value at risk if we don’t
secure this talent pool?

— How scarce is the market for this type of talent?

— How difficult is it to upskill existing or readily
available talent to fill this gap?

— How demographically diverse are the traditional
sources for this talent pool?

By clarifying which skills are most critical and which
are most easily attainable, operators can focus
investments in the areas that matter most. Over
time, they can adjust and expand into other
talent pools.

Phase three: Design tech talent strategy and
operating model
After clarifying their talent needs, telcos can begin
formulating a comprehensive talent strategy. This
should include a portfolio of innovative initiatives to
hire, train, and retain tech talent, as well as outline
the necessary infrastructure and other enablers.
Key enablers include flexible work arrangements,
learning and development platforms that function
as hubs for training resources and online courses,


enable faster data processing and secure
communications, which are particularly
attractive to business clients.

— A telco aiming to distinguish itself through
superior customer experience might prioritize
AI, which can enhance customer engagement
and customer service through personalized
offerings, automated responses, and
proactive outreach.

— A telco focusing on global reach and seamless
connectivity might prioritize ever-expanding
connectivity by investing heavily in 6G. It may
deprioritize trends like quantum technology,
which may not immediately contribute to
expanded network coverage.

— A telco seeking to increase network flexibility
while reducing costs might prioritize xRAN.
Such a telco may place less emphasis on everexpanding connectivity, as its main goal would
be improving the existing network architecture
rather than expanding reach.

— A telco looking to position itself as a technology
pioneer might prioritize quantum technology. It
may place less importance on trends like xRAN,
as its primary focus would be on pushing the
boundaries of technology rather than
restructuring the existing network.

Phase two: Assess talent gaps and define
talent priorities
Once operators are clear on the work that needs to
be done and the skills and capabilities required to
do it, they can gauge the size of the talent gaps that
will need filling and determine which types of talent
to prioritize.

They can start by mapping out current hiring and
attrition patterns against forward-looking


-----

and talent management solutions that span the
employee life cycle, from recruitment through
succession planning.

When it comes to tech talent, it is difficult to
overstate the importance of long-term thinking.
Despite recent layoffs at high-profile tech
companies, the tech talent shortage persists across
industries and could last longer than expected.
Tech unemployment in the United States is
2.1 percent—just over half the overall
unemployment rate of 3.8 percent.[4] Close to threequarters of US tech sector workers who were laid
off in 2022 found a job within three months,
according to data from Revelio Labs.[5] And the
demand for tech talent will continue to soar.

Because tech talent pipelines tend to be
particularly homogenous, telcos risk regressing on
their overall diversity goals if they fail to identify
new talent sources. In the United States, just
21 percent of those graduating with a bachelor’s
degree in computer science are women, 9 percent
are Black, and 11 percent are Latino.[6]

By thinking ahead, operators can open up the
available solution space and creatively expand their
talent pipelines into nontraditional pools and
geographical areas. Solutions like these take time
but can turn the tide—helping the sector shed its
reputation as hierarchical and stodgy, and
reposition itself as an agile, nimble, tech-forward
employer of choice.

Reimagine career development and the
employee value proposition.
As enablers of the most exciting technologies on
the horizon, operators have a powerful opportunity
to reshape how they are perceived in the talent
market. Telcos that pay close attention to tech
talent’s unique needs, desires, and priorities can
reposition themselves to attract the caliber of
talent that has seemed hopelessly out of reach
for many.

4 The tech jobs report, CompTIA, September 2023.


Recent McKinsey research shows that digital talent
places a premium on career development and
advancement potential—prioritizing these on par
with compensation. With new technologies
emerging at a dizzying pace, tech workers crave
opportunities to learn from experts and peers and
to build skills by rotating among different projects
and teams.

Our research also found that tech talent values
purpose and meaningful work. They want to
understand how the tasks that fill their own days
support the mission of the broader organization.
By creating innovative career development
opportunities and a clear sense of purpose,
telcos signal that they’re attuned to what tech
talent wants.

Build holistic university partnerships.
Across industries, leading organizations are
reimagining how they work with universities. They
are moving beyond discrete internship programs
and transactional recruiting efforts that target
graduating seniors—instead, they are establishing
durable pipelines designed with their specific talent
and diversity needs in mind. Done well, these
partnerships also provide students with highly
sought-after skills and enhance the communities in
which telcos operate.

Leading operators are building detailed models to
identify the best target universities for such
partnerships. These models assess universities’
ability to deliver large volumes of high-quality,
diverse talent. They also assess the operator’s
ability to compete with other companies and
sectors for talent at each university, based on
factors including geographical proximity, alumni
presence at the telco, network presence and
performance on campus, and ability to meet
graduates’ salary expectations.


5 Hakki Ozdenoren and Devan Rawlings, “You got laid-off. What’s next?,” Revelio Labs, December 20, 2022.

6 STEM jobs see uneven progress in increasing gender, racial, and ethnic diversity, Pew Research Center, April 1, 2021;Digest of education


-----

industries, others take a broader lens to developing
cross-industry tech talent.

Telcos may choose to start their own consortium or
join one that already exists—like the National GEM
Consortium (GEM), which recruits demographically
diverse students interested in pursuing graduatelevel degrees in applied science and engineering
and matches them with member companies in need
of their skills. GEM fellows receive stipends and
paid summer work experiences with companies
including Amazon, Meta, Ford, and Tesla.

As technology continues to evolve, so should telcos’
strategies for capturing tech talent. Early movers
give themselves the runway to experiment with
creative strategies that may pay dividends in the
long run—and to adapt and hone these strategies
based on rigorous evaluations.

Telcos’ future success rests on their ability to make
the most of the opportunities that emerging
technologies present. Multiple elements will need to
fall into place, and telco leaders are developing
ambitious transformation plans. But talent strategy
is also a critical part of the equation, and it’s often
not getting the attention it deserves. Business
leaders would be well-advised to take the reins in
shaping and steering tech talent strategy to ensure
they have the people to get the job done.


Holistic university partnerships can take different
forms. Qualcomm and several of its top executives
or directors have invested heavily over the years in
a single university, University of California San
Diego; since the late 1990s, company cofounder
Irwin Jacobs has invested more than $300 million
in the university’s engineering programs, healthcare
system, and School of Global Policy and Strategy
through scholarships and other support for
students, faculty, and research.

In another approach, Apple’s HBCU C2 initiative,
launched in partnership with Tennessee State
University, creates coding centers for learners of all
ages at historically Black colleges and universities
nationwide; it has already expanded into 45
educational institutions. And Lockheed Martin has
partnered with the University of Colorado Boulder
to fund a research center focused on radio
frequency and space systems as well as an
engineering management certificate program.

Launch or join tech talent consortiums.
Organizations are increasingly seeing the value of
partnering with businesses, government agencies,
and other players to solve collective talent
challenges. Tech talent consortiums, which provide
learners of all ages and backgrounds with skillbuilding opportunities, are promising models for
such collaboration. They may be regional or global
in nature, and while some focus on specific


Tomás Lajous is a senior partner in McKinsey’s New York office, where Stephanie Madner is an associate partner and Carlo
Palermo is a consultant; Rens van den Broek is a partner in the Bay Area office.

The authors wish to thank Davis Carlin, Vladimir Cernavskis, Zina Cole, Mena Issler, Aaron Kovar, Adam Liang, Kaitlin Noe, Katie
Owen, Caterina Priori, and Sirui Wang for their contributions to this article.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

### Deploying Gen AI


-----

People & Organizational Performance Practice, McKinsey Digital, and QuantumBlack, AI by McKinsey
###### The organization of the future: Enabled by gen AI, driven by people

Generative AI can empower people—but only if leaders take a broad view of its
capabilities and deeply consider its implications for the organization.

_by Sandra Durth, Bryan Hancock, Dana Maor, and Alexander Sukharevsky_

September 2023


-----

_“We were behind on automation and digitization,_
_and we finally closed the gap. We don’t want to be_
_left behind again, but we aren’t sure how to think_
_about generative AI.”_

That’s the sentiment shared by many global
executives, given the speed with which generative
artificial intelligence (gen AI)[1] is advancing in the
business world. The technology is accessible,
ubiquitous, and promises to have a significant
impact on organizations and the economy over the
next decade.

Anyone can use gen AI, with little or no formal
training or technical know-how. It is being
embedded in everyday tools, like email, word
processing applications, and meeting software,
which means the technology is already positioned
to radically transform how people work. And
McKinsey research shows that gen AI could enable
automation of up to 70 percent of business
activities, across almost all occupations, between
now and 2030, adding trillions of dollars in value to
the global economy.[2]

Meanwhile, technologists keep reminding us that
gen AI is only in its nascent stages of development
and usage. This smart technology is only going to
get more intelligent—and those who don’t learn to
work with it, starting now, will be left behind.[3]

In this supercharged environment, how can
organizations do more than just “keep up”? What
strategies, structures, and talent management
approaches will business leaders need to adopt to
prepare their organizations for a gen-AI-driven
future? We examine these and other critical
questions in this article.

The situation is evolving rapidly, and there is,
frankly, no one right answer to the question of how
to successfully roll out gen AI in the organization—
business context matters.


But to start, business leaders will need to think
broadly about how the rollout of Gen AI could affect
their organizations day to day—especially their
people. Employees and managers should have a
clear understanding of gen AI’s strengths and
weaknesses and how use of the technology is
linked to the organization’s strategic objectives.
Given the technology’s potential to accelerate
automation, senior leaders could counter
employees’ prevailing fears of “replacement and
loss” with messaging about gen AI’s potential for
“augmentation and improvement”—and its ability to
significantly enhance the employee experience.
Imagine, for example, a world with fewer meetings
and more time to think.

The central task for senior leaders, then, is to
demystify the technology for others; that will mean
taking a step back to assess the strategic
implications of gen AI, or the risks and opportunities
for industries and business models. As leaders build
a compelling narrative for the use of gen AI, they
will also need to identify two or three high-impact
applications to explore and bring employees along
on a value-creating journey—taking gen AI
initiatives from pilot test to rapid scaling to “business
as usual” status. Senior leaders will also need to
commit to building the required roles, skills, and
capabilities (now and for the future), so they can
continually test and learn with gen AI and stay
ahead of competitors.

**Are you thinking broadly enough**
**about gen AI’s potential impact?**

McKinsey research suggests that, because of the
emergence of gen AI, about half of today’s business
activities could be automated a decade earlier than
previous estimates had projected.[4] Gen-AI-enabled
automation has already begun—and, as the
research shows, is likely to affect hours, tasks, and
responsibilities for workers across wage rates and


1 Generative AI is a form of AI that can generate text, images, or other content in response to user prompts. It differs from previous generations of
AI, in part, because of the scope of outputs it can create.

2 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.

3 Paolo Confino and Amber Burton, “A.I. might not replace you, but a person who uses A.I. could,” Fortune, April 25, 2023.


-----

educational backgrounds. In fact, the research
shows that gen AI will have an especially profound
effect on professions traditionally requiring higher
levels of education, such as educators and lawyers.[5]

Gen AI is also likely to inform discussions in the
C-suite about how the company creates value and
whether the addition of gen AI capabilities allows
for industry or business model reinvention. As a
result, leaders should ask themselves a range of
critical questions relating to the “new” nature of
work in gen-AI-enabled organizations, including
the following:

_What are the organization-wide implications of_
_gen AI?_ Rather than taking a passive approach to
identifying potential use cases and investments
associated with gen AI, leaders should view the
situation through an “attacker’s lens.” They should
consider all the primary, secondary, and even
tertiary effects of gen AI: Which business use cases
are highest priority now—and which might be
candidates for gen AI enablement in six months, 12
months, and so on? What changes will be required
at a functional level to make gen AI enablement
possible—for instance, how many more software
engineers will the company need? And as gen AI
functionality continues to be embedded in common
word processing, email, and communications tools
(Microsoft’s 365 Copilot, for instance), what effect
will that have on ways of working across the entire
organization? Could gen AI accelerate the shift to a
four-day work week? And even more broadly, how
might entire industries or business models be
fundamentally disrupted?

_Does the organization have the right technical_
_talent and risk infrastructure in place?_ Leaders
should consider which operating-model designs will
be most effective for ensuring the long-term
development of technology talent and the


continued evolution of gen AI applications in the
organization (see sidebar “Speeding up the search
for tech talent”). They should also consider whether
that same structure can satisfy the need for gen AI
oversight (see sidebar “A powerful resource with
potential risks”).

_How can corporate culture enable or inhibit the_
_adoption and usage of gen AI?_ Gen AI applications
can be the catalyst for culture change—in more
ways than one. The applications themselves can
create more organizational transparency and
connectivity. One company, for instance, is piloting
a gen AI application that allows users to ask
questions about operations, sales, and other topics,
and the tool draws from the company’s entire
collection of intellectual property to come up with
answers that can guide users to the most relevant
experts and data. Employees report feeling better
informed and more connected. Additionally, the
same cultural traits that have been crucial for
organizational success during recent economic and
business upheavals—such as adaptability, speed,
agility, trust, integrity, learning and experimentation,
innovation, and a willingness to change—will be
even more important if organizations want to
become truly enabled by gen AI. To understand
why, consider the findings from the 2023 McKinsey
Digital survey of 1,000 organizations, which found a
significant synergy between organizations with
strong, innovative cultures and their ability to
increase value through new digital technologies,
including gen AI.[6] In previous iterations of that
survey, respondents said the biggest obstacle to
their digital success was a culture that was averse
to risk and experimentation.[7]

_How should organizations change their talent_
_management approaches?_ Gen AI applications will
have unprecedented effects on organizations’
approaches to talent management. Consider the


5 “Generative AI and the future of work in America,” McKinsey, July 26, 2023.

6 “Companies with innovative cultures have a big edge with generative AI,” McKinsey, August 31, 2023


-----

**Speeding up the search for tech talent**


In the coming months and years,
demand for those who have mastered
working with and alongside gen AI will
skyrocket—especially for those who build
and engineer gen AI tools and those who
are in the business of generating content
via gen AI. (We call the latter “creators,”
and they can include product managers,
marketing managers, and so on.)

To speed up and simplify the search for
this critical tech talent amid heavy
competition, business leaders should first
identify the types of gen AI applications
they need to build. They can then use
those insights to identify the type and
amount of tech talent they will need in the
short term—and how to retain that talent
for the longer term.

_What gen AI applications are we building_
_ourselves?_ The first decision involves
deciding—in collaboration with IT, R&D,
and business unit leaders—what


applications to build and what applications
to adapt from off-the-shelf products.[1] Gen
AI applications can be expensive and
complicated to build, requiring significant
technical know-how. Once built, the
applications must be continuously updated
or risk losing utility. What’s more, training
new gen AI applications takes significantly
more energy than using or refining
existing ones.

_Who do we need to build these gen AI_
_applications? Once they know what_
applications they need to build and buy,
senior leaders can examine the technology
roles and responsibilities they will need to
create value from gen AI. Organizations will
need engineering and software
development talent, but they will also need
translator roles—including implementation
coaches, educators, and trainers—to
facilitate the understanding and adoption
of gen AI across the organization.


_How do we develop and retain this tech_
_talent?_ According to McKinsey research,
opportunities for career development, the
potential for advancement, and
compensation are the top factors
technology professionals consider. The
chance to learn is another key draw, with
professionals reporting a desire to work in
an organization that provides employees
with opportunities to practice new skills.[2]
To meet these requirements and increase
the likelihood of retaining top tech talent,
senior leaders could explore the use of
programs such as peer-to-peer learning,
functional rotations that expose
technologists to other parts of the
organization, and upskilling.[3]


1 “What every CEO should know about generative AI,” McKinsey, May 12, 2023.

2 “Cracking the code on digital talent,” McKinsey, April 20, 2023.

3 Vincent Bérubé, Dana Maor, Maria Ocampo, and Alex Sukharevsky, “HR rewired: An end-to-end approach to attracting and retaining top tech talent,” McKinsey,

June 27, 2023.


inevitable impact of gen AI applications on
apprenticeship, particularly in the case of
knowledge work: imagine a marketing leader uses a
gen AI application to write a creative brief that
previously would have been developed by a more
junior marketing associate. What will happen to the


development and mentorship opportunities for both
the leader and associate when the learning process
is disintermediated by gen AI? What’s more, both
the content and the delivery of skill-building
programs will be affected. A chatbot could guide


-----

**A powerful resource with potential risks**

**A powerful resource with potential risksBefore business leaders can successfully** reliability of gen AI models, which can model to create disinformation, deepfakes,

incorporate gen AI into their business produce different answers to the same and other types of malicious content.
strategies and organizations, they must be prompts and present “hallucinations” as

Organizations will need to take a proactive

Before business leadersclear about the risks it may pose and can successfully reliability of gen AI models, which can compelling facts. model to create disinformation, deepfakes,

role in educating regulators about the

incorporate gen AI into their business anticipate potential responses; it’s the only produce different answers to the same and other types of malicious content.

Organizations may have trouble shielding business uses of gen AI and engaging with

strategies and organizations, they must be way to maintain trust with and among prompts and present “hallucinations” as

some of their intellectual property Organizations will need to take a proactive standards bodies to ensure a safe and

clear about the risks it may pose and employees, investors, and customers.[1]compelling facts.

(copyrights, trademarks, patents, and other role in educating regulators about the competitive future with the technology.

anticipate potential responses; it’s the only

Among the risks are concerns about the Organizations may have trouble shielding legally protected materials) from being business uses of gen AI and engaging with

way to maintain trust with and among

types of biases that may be built into gen some of their intellectual property inadvertently exposed through a standards bodies to ensure a safe and

employees, investors, and customers.[1]

AI applications, which could negatively (copyrights, trademarks, patents, and other company’s gen AI outputs. Similarly, bad competitive future with the technology.

Among the risks are concerns about the affect specific groups in an organization. legally protected materials) from being actors could plug sensitive customer,
types of biases that may be built into gen There may also be questions about the inadvertently exposed through a supplier, or employee data into a gen AI
AI applications, which could negatively company’s gen AI outputs. Similarly, bad
affect specific groups in an organization. actors could plug sensitive customer,
There may also be questions about the 1 Jim Boehm, Liz Grennan, Alex Singla, and Kate Smaje,supplier, or employee data into a gen AI “Why digital trust truly matters,” McKinsey, September 12, 2022.


1 Jim Boehm, Liz Grennan, Alex Singla, and Kate Smaje, “Why digital trust truly matters,” McKinsey, September 12, 2022.

new employees through training on a new at the center.[9] Work processes should enable
technology, at their own pace, on their own terms, short, quick cycles of experimentation and iteration
allowing them to increase the extent and speed of and high-quality feedback loops among employees,
their learning.[8] Meanwhile, their instructor may use leaders, and the gen AI applications themselves. To

new employees through training on a new at the center.[9] Work processes should enable

a gen-AI-enabled “teaching assistant” app to create that end, it can be helpful to build small cross
technology, at their own pace, on their own terms, short, quick cycles of experimentation and iteration

engaging training modules for individuals and functional teams working end to end on projects

allowing them to increase the extent and speed of and high-quality feedback loops among employees,

groups and to track the progress of both. and initiatives.

their learning.[8] Meanwhile, their instructor may use leaders, and the gen AI applications themselves. To
a gen-AI-enabled “teaching assistant” app to create that end, it can be helpful to build small cross
These are just a few key organizational

engaging training modules for individuals and considerations; many more are still evolving. functional teams working end to end on projects People and gen AI: Building an
groups and to track the progress of both.Decisions on structure and operating-model design, and initiatives.empowered workforce

for instance, will vary from company to company, Gen AI can be a powerful tool for employee

These are just a few key organizational
considerations; many more are still evolving. but whatever the form, our decades-long **People and gen AI: Building an empowerment—even among those who initially**
Decisions on structure and operating-model design, experience with digital transformations suggests empowered workforceperceive it as a threat:

that discussions about value creation must remain

for instance, will vary from company to company, Gen AI can be a powerful tool for employee
but whatever the form, our decades-long empowerment—even among those who initially
experience with digital transformations suggests perceive it as a threat:
that discussions about value creation must remain


8 Benjamin S. Bloom, “The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring.” Educational
Researcher, June–July 1984, Volume 13, Number 66.

9 Eric Lamarre, Kate Smaje, and Rodney Zemmel, Rewired: The McKinsey Guide to Outcompeting in the Age of Digital and AI, New York,
NY: [Wiley, 2023.]

8 Benjamin S. Bloom, “The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring.” Educational The organization of the future: Enabled by gen AI, driven by people
Researcher, June–July 1984, Volume 13, Number 66.

9 Eric Lamarre Kate Smaje and Rodney Zemmel Rewired: The McKinsey Guide to Outcompeting in the Age of Digital and AI New York


-----

groups; research suggests this work could
dramatically increase the number and diversity of
applications for various roles.[12] Gen AI applications
could also help companies match new hires with
mentors and coaches to improve the onboarding
experience, upskill talent, and streamline
administrative tasks.

It can prompt senior leaders to lead differently.
Senior leaders face the dual responsibility of quickly
implementing gen AI today and anticipating future
versions of gen AI technologies and their
implications. More than anyone else in the
organization, they will need to be evangelists for gen
AI, encouraging the development and adoption of
the technology organization wide. That will mean
working with other business unit and technology
leaders to allocate resources to update technology
infrastructure and take any interim process steps
required to facilitate the gen AI rollout—for instance,
moving applications to private cloud-hosted
environments. In fact, a central task for senior
leaders will be to find ways to forge stronger
connections between technology leaders and the
business units. One company, for example, launched
a Slack channel devoted to ongoing discussion of
gen AI pilots. Through such forums, employees,
product developers, and other business and
technology leaders can share stories about their
experiences with gen AI, whether and how their
daily tasks have changed, and their thoughts on the
gen AI journey so far.

As they would when introducing any new
technology, senior leaders should speak clearly
about the business objectives of gen AI,
communicating early and often about gen AI’s role
in “augmenting versus replacing” jobs. They should
paint a compelling picture of how various aspects of
the organization will be rewired through gen AI—
technically, financially, culturally, and so on.


It can augment the employee experience. Gen AI
applications can assist employees in ways that
many workers may not even expect. For instance,
gen AI can suggest the new lines of code required
to update a financial-reporting system or outline
the A and B versions of a marketing campaign or
otherwise create first drafts that human employees
can take and implement in live production
environments. And by facilitating the training and
upskilling process, gen AI applications can help
employees pick up new skills more quickly. A recent
study, for instance, found that software engineers
completed their coding tasks up to twice as fast
when using gen AI and reported more satisfaction
with the process.[10]

It can empower middle managers. The benefits of
gen AI can accrue not just to frontline workers but
also to middle managers. In fact, as the people
closest to employees, middle managers have a
critical role to play in increasing employees’ comfort
with both short-term gen-AI-enabled work and
long-term collaborations with the technology.[11] And
as their own direct reports learn to work with gen
AI, middle managers may find themselves
overseeing more and different kinds of work
streams, moving at a pace never seen before. At
the same time, the use of gen AI can free up
more capacity for middle managers, so they can
shift their attention to higher-value leadership
tasks, such as strategy-focused work and
people management.

It can help organizations reinvent their talent
management practices. The emergence of gen AI
presents an opportunity for organizations to hone
their approaches to attracting, retaining, and
developing talent—particularly when it comes to
creators and tech professionals. HR professionals
could use gen AI to send personalized outreach
emails to candidates and to design job search
experiences for candidates in underrepresented


10 “Unleashing developer productivity with generative AI,” McKinsey, June 27, 2023.

11 People & Organization Blog, blog post by Emily Field, Bryan Hancock, Ruth Imose, and Lareina Yee, “Middle managers hold the key to unlock
generative AI,” McKinsey, July 19, 2023.

12 Justin Friesen, Danielle Gaucher, and Aaron C. Kay. “Evidence that gendered wording in job advertisements exists and sustains gender


-----

big” on those that show the greatest promise of
scalability and long-term value—whether it’s an
application that simplifies financial reporting or one
that enhances onboarding for new hires. As part of
this vetting process, senior leaders should consider
the business or industry risks or opportunities
associated with implementing the gen AI pilot, as
well as how hard or easy it will be to move the pilot
into production and make it a part of employees’
day-to-day experiences. Once that vetting has
happened, senior leaders should steer resources
accordingly and take care to monitor and measure
the outputs from gen AI initiatives and pilots.
Remember, some gen AI initiatives may show
impact in the next 12 months, while others may
require investment now to yield results in two to
five years. The longer-term goal, then, should be to
set up a sustainable engine for the rapid upskilling
of employees and scaling of gen AI and other
digital capabilities.

Commit to building the necessary roles, skills,
and capabilities—now and in the future. Senior
leaders should commit to building employees’ gen
AI skills so they can use the technology judiciously
and successfully in their day-to-day work. It’s not a
one-and-done process; leaders will need to
continually assess how and when tasks are
performed, who is performing them, how long tasks
typically take, and how critical different tasks are.
Through this process, leaders can better
understand current and future talent needs and
determine how best to redeploy and upskill talent.
Indeed, upskilling programs will take on greater
importance than ever, as employees will need to
learn to manage and work with gen AI tools that are
themselves ever evolving. Leaders should also keep
in mind that gen AI itself may facilitate the creation
of content for, and automated or personalized
delivery of, such upskilling programs.


Of course, if senior leaders don’t understand the
technology themselves, it will be more difficult to
make this case for, and lead their teams into, a genAI-enabled future. One way for leaders to stay
plugged in is to establish forums that provide
ongoing professional education on advances in AI
technology and applications. Another approach is to
carve out time during planning meetings to
consider forward-looking questions such as, “Is our
approach to gen AI today flexible enough to
accommodate the next iteration, and the one after
that?” and “Which process steps or roles will we be
able to reinvent with the next iteration of gen AI?”

**Time to flex your gen AI muscle**

Although generative AI burst onto the scene
seemingly overnight, CEOs and other business
leaders can ill afford to take an overly cautious
approach to introducing it in their organizations. If
ever a business opportunity demanded a bias for
action, this is it. By taking the following three steps
simultaneously, and with a sense of urgency,
leaders can do more than just “keep up”—they can
capture early gains and stay ahead of competitors.

Demystify gen AI for everyone. Senior leaders
themselves should develop a deep understanding
of gen AI and associated capabilities themselves so
they can help to demystify the technology for the
rest of the organization. They can then help to
introduce mechanisms for managing uncertainties
about gen AI where they exist—for instance,
establishing clear guidance regarding the use of
gen AI tools in hiring and recruiting where AI model
biases could emerge.

Identify two or three high-impact use cases—and
just get started. Senior leaders should carefully
consider their investments in gen AI pilots, and “go


-----

In the time it took to read this article, gen AI
applications have already gotten that much smarter.
Leaders can put that intelligence to good use. It’s
clear that much of the value of gen AI will come
from tailoring it to organization-specific use


cases—but the successful integration of gen AI
requires experimentation and iteration. There is no
time to sit back and learn from others’ mistakes.
Invest deliberately. Get your hands dirty. Start now.


Sandra Durth is an associate partner in McKinsey’s Cologne office, Bryan Hancock is a partner in the Washington, DC, office,
Dana Maor is a senior partner in the Tel Aviv office, and Alexander Sukharevsky is a senior partner in the London office.

The authors wish to thank Jan Bouly, Michael Chui, Neel Gandhi, Randy Lim, Federico Marafante, Maria Ocampo, Joachim
Talloen, Alon Van Dam, and Anna Wiesinger for their contributions to this article.

This article was edited by Roberta Fusaro, an editorial director in the Waltham, Massachusetts, office.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

**The data dividend: Fueling**
**generative AI**

Data leaders should consider seven actions to enable companies to scale
their generative AI ambitions.

_This article is a collaborative effort by Joe Caserta, Holger Harreis, Kayvaun Rowshankish, Nikhil Srinidhi,_
_and Asin Tavakoli, representing views from McKinsey Digital._


© Getty Images


-----

If your data isn’t ready for generative AI, your
business isn’t ready for generative AI.

Our latest research estimates that generative
AI could add the equivalent of $2.6 trillion to
$4.4 trillion in annual economic benefits across
63 use cases.¹ Pull the thread on each of these
cases, and it will lead back to data. Your data and
its underlying foundations are the determining
factors to what’s possible with generative AI.

That’s a sobering proposition for most chief data
officers (CDOs), especially when 72 percent of
leading organizations note that managing data
is already one of the top challenges preventing
them from scaling AI use cases.² The challenge
for today’s CDOs and data leaders is to focus
on the changes that can enable generative AI to
generate the greatest value for the business.

The landscape is still rapidly shifting, and there
are few certain answers. But in our work with
more than a dozen clients on large generative
AI data programs, discussions with about 25
data leaders at major companies, and our own
experiments in reconfiguring data to power
generative AI solutions, we have identified seven
actions that data leaders should consider as they
move from experimentation to scale:

1. Let value be your guide. CDOs need to be
clear about where the value is and what data
is needed to deliver it.

2. Build specific capabilities into the data
architecture to support the broadest set
of use cases. Build relevant capabilities
(such as vector databases and data pre- and
post-processing pipelines) into the existing
data architecture, particularly in support of
unstructured data.

3. Focus on key points of the data life cycle
to ensure high quality. Develop multiple
interventions—both human and automated—
into the data life cycle from source to


consumption to ensure the quality of all material
data, including unstructured data.

4. Protect your sensitive data, and be ready to
move quickly as regulations emerge. Focus on
securing the enterprise’s proprietary data and
protecting personal information while actively
monitoring a fluid regulatory environment.

5. Build up data engineering talent. Focus on
finding the handful of people who are critical to
implementing your data program, with a shift
toward more data engineers and fewer data
scientists.

6. Use generative AI to help you manage your
own data. Generative AI can accelerate existing
tasks and improve how they’re done along the
entire data value chain, from data engineering to
data governance and data analysis.

7. Track rigorously and intervene quickly. Invest
in performance and financial measurement, and
closely monitor implementations to continuously
improve data performance.

**1. Let value be your guide**
In determining a data strategy for generative
AI, CDOs might consider adapting a quote from
President John F. Kennedy: “Ask not what your
business can do for generative AI; ask what
generative AI can do for your business.” Focus on
value is a long-standing principle, but CDOs must
particularly rely on it to counterbalance the pressure
to “do something” with generative AI. To provide
this focus on value, CDOs will need to develop a
clear view of the data implications of the business’s
overall approach to generative AI, which will play out
across three archetypes:

— Taker: a business that consumes preexisting
services through basic interfaces such as APIs.
In this case, the CDO will need to focus on
making quality data available for generative AI
models and subsequently validating the outputs.


1 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.
2 McKinsey Data & AI Summit 2022.


-----

capabilities. One key function in driving the Shaper
approach is communicating the trade-offs needed to
deliver on specific use cases and highlighting those
that are most feasible. While hyperpersonalization,
for example, is a promising generative AI use case, it
requires clean customer data, strong guardrails for
data protection, and pipelines to access multiple data
sources. The CDO should also prioritize initiatives that
can provide the broadest benefits to the business,
rather than simply support individual use cases.

As CDOs help shape the business’s approach to
generative AI, it will be important to take a broad
view on value. As promising as generative AI is, it’s
just one part of the broader data portfolio (Exhibit 1).
Much of the potential value to a business comes from
traditional AI, business intelligence, and machine
learning (ML). If CDOs find themselves spending


— Shaper: a business that accesses models and
fine-tunes them on its own data. The CDO
will need to assess how the business’s data
management needs to evolve and what changes
to the data architecture are needed to enable
the desired outputs.

— Maker: a business that builds its own
foundational models. The CDO will need
to develop a sophisticated data labeling
and tagging strategy, as well as make more
significant investments.

The CDO has the biggest role to play in supporting
the Shaper approach, since the Maker approach
is currently limited to only those large companies
willing to make major investments and the Taker
approach essentially accesses commoditized


Exhibit 1
**Take a portfolio view on value.**

Take a portfolio view on value.

Illustrative banking use cases portfolio

High

Impact Medium

Low

|Hyperpersonalized marketing campaigns|Voice bot Churn reduction Next-product-to-buy algorithms|Optimization of credit underwriting Marketing budget allocation Code development/ maintenance productivity Customer operations|
|---|---|---|
|Fraud detection Frontline customer copilot|Gen AI–based designs and content Risk report generation|Wealth customer profling Call analysis Virtual customer-service agent ATM locations optimization|
|Regulatory fling and drafting Tailored automated onboarding and training|Talent screening Automated accounting from receipts|Text-based alerts Order-entry assistant Discovery of regulatory changes Customer behavior analytics|


Impact


Low

Feasibility


M Ki & C


Generative AI Business intelligence and analytics Classical AI/ML


Medium

Feasibility


High


-----

90 percent of their time on initiatives related to
generative AI, that’s a red flag.

**2. Build specific capabilities into**
**the data architecture to support the**
**broadest set of use cases**
The big change when it comes to data is that the
scope of value has gotten much bigger because
of generative AI’s ability to work with unstructured
data, such as chats, videos, and code. This
represents a significant shift because data
organizations have traditionally had capabilities
to work with only structured data, such as data
in tables. Capturing this value doesn’t require a
rebuild of the data architecture, but the CDO who
wants to move beyond the basic Taker archetype
will need to focus on two clear priorities.

The first is to fix the data architecture’s
foundations. While this might sound like old
news, the cracks in the system a business could
get away with before will become big problems
with generative AI. Many of the advantages of
generative AI will simply not be possible without a
strong data foundation. To determine the elements
of the data architecture on which to focus, the
CDO is best served by identifying the fixes that
provide the greatest benefit to the widest range
of use cases, such as data-handling protocols for
personally identifiable information (PII), since any
customer-specific generative AI use case will need
that capability.

The second priority is to determine which upgrades
to the data architecture are needed to fulfill the
requirements of high-value use cases. The key
issue here is how to cost-effectively manage
and scale the data and information integrations
that power generative AI use cases. If they are
not properly managed, there is a significant
risk of overstressing the system with massive
data compute activities, or of teams doing oneoff integrations, which increase complexity
and technical debt. These issues are further
complicated by the business’s cloud profile, which
means CDOs must work closely with IT leadership
to determine compute, networking, and service
use costs.


In general, the CDO will need to prioritize the
implementation of five key components of the
data architecture as part of the enterprise tech
stack (Exhibit 2):

— Unstructured data stores: Large language
models (LLMs) primarily work with
unstructured data for most use cases. Data
leaders will need to map out all unstructured
data sources and establish metadata tagging
standards so models can process the data
and teams can find the data they need. CDOs
will need to further upgrade the quality of
data pipelines and establish standards for
transparency so that it’s easy to track the
source of an issue to the right data source.

— Data preprocessing: Most data will
need to be prepped—for example, by
converting file formats and cleansing for
data quality and the handling of sensitive
data—so that generative AI can use the
data. Preprocessed data is most often used
to build prompts for generative AI models.
To speed up performance, CDOs need to
standardize the handling of structured and
unstructured data at scale, such as ways to
access underlying systems, and prioritize (or
“preaggregate”) the data that supports the
most frequent questions and answers.

— Vector databases: Vectorization is a way to
prioritize content and create “embeddings”
(numerical representations of text meanings)
in order to streamline access to context, the
complementary information generative AI
needs to provide accurate answers. Vector
databases allow generative AI models to
access just the most relevant information.
Instead of providing a thousand-page PDF,
for example, a vector database provides
only the most relevant pages. In many
cases, companies don’t need to build vector
databases to begin working with generative
AI. They can often use existing NoSQL
databases to start.

— LLM integrations: More-sophisticated
generative AI uses require interactions with


-----

Exhibit 2 Upgrades are needed within the existing data architecture to enable
**Upgrades are needed within the existing data architecure to enable generative AI.generative AI.**


Illustrative data architecture

Data and model governance


Gen AI exensions, with
mature tooling/solutions


Gen AI exensions, with
novel/emerging tooling/solutions

|Data sources|Data ingestion Batch data integration Event streaming|Data repositories Rational Unstructured data and database metadata stores Graph database Vector database Document- (chunking, indexing, oriented and creating database embeddings)|Col4|Data services Data API endpoints and API management Access data (structured and unstructured data) Prompt engineering • Integrate endpoints of data model ontologies and knowledge graphs • Remove PII information (if not done during preprocessing) • Perform data retrieval to include in prompt • Execute similarity search against vector database|Data consumption Advanced analytics Business intelligence and report- ing Gen AI application|
|---|---|---|---|---|---|
|Structured data sources||||||
|Unstructured data sources||||||
|||||||
|Processing||||||
|Stream processing|Gen AI preprocessing • Preaggregate data for answering questions (eg, prioritize data that support the most frequent questions and answers) • Prepare data to feed into LLM (eg, fle-format conversion, cleansing for data quality, and sensitive data handling)||LLMs (closed source, open source, and/or private)|||
|Batch processing||||||
|AI/ML||||||


|MDM¹|Col2|Data governance: data model ontologies, data transparency and quality, access policies, data product cards, data usage by gen AI, data tagging|Col4|Col5|Col6|AI model governance: model transparency, outcome monitoring, model shift|
|---|---|---|---|---|---|---|
|ML model governance|||||||
|Control center “gateway”|||||||
|DataOps|MLOps/LLMOps||LiveOps|FinOps|LLM gateway (trafc monitoring, request logging, credential management, FinOps, model access, PII protection)||


1Master data management.

McKinsey & Company

multiple systems, which creates significant
challenges in connecting LLMs. Several
frameworks, many of which are open source,
can help facilitate these integrations (for
example, LangChain or various hyperscaler
offerings, such as Semantic Kernel for Azure,
Bedrock for AWS, or Vertex AI for Google
Cloud). CDOs will need to set guidelines
for choosing which frameworks to use,
define prompt templates that can be readily
customized for specific purposes, and establish
standardized integration patterns for how
LLMs interface with source data systems.


— Prompt engineering: Effective prompt
engineering (the process of structuring
questions in a way that elicits the best response
from generative AI models) relies on context.
Context can be determined only from existing
data and information across structured and
unstructured sources. To improve output, CDOs
will need to manage integration of knowledge
graphs or data models and ontologies (a set of
concepts in a domain that shows their properties
and the relations between them) into the prompt.
Since CDOs will not have ownership of many
data repositories across the business, they


-----

— Prompt: Evaluate, measure, and track the
quality of the prompt. Include high-quality
metadata and lineage transparency for
structured and unstructured data in the prompt.

— Output from LLM: Establish the necessary
governance procedures to identify and resolve
incorrect outputs, and use “human in the loop”
to review and triage output issues. Ultimately,
elevate the role of individual employees by
training them to critically evaluate model
outputs and be aware of the quality of
input data. Supplement with an automated
monitoring-and-alert capability to identify
rogue behaviors.

**4. Protect your sensitive data, and be**
**ready to move quickly as regulations**
**emerge**
Some 71 percent of senior IT leaders believe
generative AI technology is introducing new
security risk to their data.⁵ Much has been
written about security and risk when it comes to
generative AI, but CDOs needs to consider the
data implications in three specific areas:

— Identify and prioritize security risks to the
enterprise‘s proprietary data. CDOs need
to assess the broad risks associated with
exposing the business’s data, such as the
potential exposure of trade secrets when
confidential and proprietary code is shared
with generative AI models, and prioritize
the greatest threats. Much existing data
protection and cybersecurity governance can
be extended to address specific generative
AI risks—for example, by adding pop-up
reminders whenever an engineer wants
to share data with a model or by running
automated scripts to ensure compliance.

— Manage access to PII data. CDOs need to
regulate how data is detected and treated
in the context of generative AI. They need to


need to set standards and prequalify sources
to ensure the data that is fed into the models
follows specific protocols (for example, exposing
a knowledge graph API to easily provide entities
and relationships).

**3. Focus on key points of the data life**
**cycle to ensure high quality**
Data quality has always been an important issue
for CDOs. But the scale and scope of data that
generative AI models rely on has made the “garbage
in/garbage out” truism much more consequential
and expensive, as training a single LLM can cost
millions of dollars.³ One reason pinpointing data
quality issues is much more difficult in generative
AI models than in classical ML models is because
there’s so much more data and much of it is
unstructured, making it difficult to use existing
tracking tools.

CDOs need to do two things to ensure data quality:
extend their data observability programs⁴ for
generative AI applications to better spot quality
issues, such as by setting minimum thresholds for
unstructured content to be included in generative AI
applications; and develop interventions across the
data life cycle to fix the issues teams find, mainly in
four areas:

— Source data: Expand the data quality
framework to include measures relevant for
generative AI purposes (such as bias). Ensure
high-quality metadata and labels for structured
and unstructured data, and regulate access
to sensitive data (for example, base access on
roles).

— Preprocessing: Ensure data is consistent
and standardized and adheres to ontologies
and established data models. Detect outliers
and apply normalizations. Automate PII data
management, and put in place guidelines for
whether data should be ignored, held, redacted,
quarantined, removed, masked, or synthesized.


3 B. Urian, “NVIDIA announces $9.6 million drop in cost when using its GPUs for AI LLM training,” Tech Times, May 29, 2023.
4 Data observability programs consist of mechanisms for understanding the health and performance of the data within systems.
5 “Top generative AI statistics for 2023,” Salesforce, September 2023.


-----

set up systems that incorporate protection
tools and human interventions to ensure PII
data is removed during data preprocessing
and before it’s used on an LLM. Using
synthetic data (through data fabricators) and
nonsensitive identifiers can help.

— Track the expected surge of regulations
closely. Generative AI has acted as a catalyst
to rapid movement among governments to
enact new regulations, such as the European
Union’s AI Act, which is setting a wide array
of new standards, such as having companies
publish summaries of copyrighted data
used for training an LLM. Data leaders must
stay close to the business’s risk leaders
to understand new regulations and their
implications for data strategy, such as the
need to “untrain” models that use regulated
data.

**5. Build up data engineering talent**
As enterprises increasingly adopt generative
AI, CDOs will have to focus on the implications
for talent. Some coding tasks will be done by
generative AI tools—41 percent of code published
on GitHub is written by AI.⁶ This requires
specific training on working with a generative AI
“copilot”—a recent McKinsey study showed that
senior engineers work more productively with a
generative AI copilot than do junior engineers.⁷
Data and AI academies need to incorporate
generative AI training tailored to specific
expertise levels.

CDOs will also need to be clear about what skills
best enable generative AI. Companies need
people who can integrate data sets (such as
writing APIs connecting models to data sources),
sequence and chain prompts, wrangle large
quantities of data, apply LLMs, and work with
model parameters. This means that CDOs should
focus more on finding data engineers, architects,
and back-end engineers, and less on hiring


data scientists, whose skills will be increasingly
less critical as generative AI allows people with
less advanced technical capabilities to use natural
language in doing basic analysis.

In the near term, talent will remain in shorter supply,
and we project that the talent gap will increase
further in the near future,⁸ creating more incentives
for CDOs to build up their training programs.

**6. Use generative AI to help you**
**manage your own data**
Data leaders have a huge opportunity to harness
generative AI to improve their own function. In our
analysis, eight primary use cases have emerged
along the entire data value chain where generative
AI can both accelerate existing tasks and improve
how tasks are performed (Exhibit 3).

Many vendors are already rolling out products,
requiring CDOs to identify the capabilities for which
they can rely on vendors and which they should
build themselves. One rule of thumb is that for
data governance processes that are unique to the
business, it’s better to build your own tool. Note that
many tools and capabilities are new and may work
well in experimental environments but not at scale.

**7. Track rigorously and intervene**
**quickly**
There are more unknowns than knowns in the
generative AI world today, and companies are still
learning their way forward. It is therefore crucial
for CDOs to set up systems to actively track and
manage progress on their generative AI initiatives
and to understand how well data is performing in
supporting the business’s goals.

In practice, effective metrics are made up of a set
of core KPIs and operational KPIs (the underlying
activities that drive KPIs), which help leaders track
progress and identify root causes of issues.


6 Jose Antonio Lanz, “Stability AI CEO: There will be no (human) programmers in five years,” Decrypt, June 3, 2023.
7 “Unleashing developer productivity with generative AI,“ McKinsey, June 27, 2023.
8 Michael Chui, Mena Issler, Roger Roberts, and Lareina Yee, “McKinsey Technology Trends Outlook 2023,” McKinsey, July 20, 2023.


-----

Exhibit 3
**Generative AI opportunities exist to improve the entire date value chain.Generative AI opportunities exist to improve the entire data value chain.**

Generative AI use cases along data value chain

1. Data discovery
Executes queries to build a data profle of each major data repository and makes it interactive for users

2. Data ingestion
Creates data-ingestion templates with connection strings and ports, reflls/creates tokens, and readies
the data for deployment to abstract the technical system away from data engineers

3. Data storage and curation
Uses intelligent data compression and archiving algorithms to optimize storage capacity and reduce costs
(based on usage logs)

4. Data processing
Determines best computing frameworks for processing large volumes of data in real time (based on use
case profle)

5. Data access
Determines access rules and who should and shouldn’t have access to what type of data by analyzing
policies and active directory metadata

6. Data consumption
Integrates with business intelligence and reporting tools for improved data storytelling and
collaboration (copilot)

7. Data governance
Integrates with data privacy and protection solutions, such as data masking and encryption, and provides
real-time monitoring and alerts for data breaches, anomalous behavior, and other security threats

8. Data interpretation
Improves data interpretation by providing recommendations based on the insights generated from
the data

McKinsey & Company


A core set of KPIs should include the following:

— cost of additional components, such as vector
databases and consumption of LLMs as a
service

— additional revenue that is enabled by the
integration of specific data sources with
generative AI application workflows

— time-to-market to develop a generative AI–
powered application that requires access to
internal data

— end-user satisfaction with how the data has
improved the performance and quality of the
application


Operational KPIs should include tracking which data
is being used most, how models are performing,
where data quality is poor, how many requests are
being made against a given dataset, and which use
cases are generating the most activity and value.

This information is critical in providing a fact base for
leadership to not just track progress but also make
rapid adjustments and trade-off decisions against
other initiatives in the CDO’s broader portfolio. By
knowing which data sources are most used for highvalue models, for example, the CDO can prioritize
investments to improve data quality at those
sources.


-----

Effective investment, budgeting, and reallocation
will depend on CDOs developing a FinOpslike capability to manage the entire new cost
structure growing around generative AI. CDOs
will need to track a new range of costs, including
the number of generative AI model requests,
API consumption charges from vendors (both
quantity and size of calls), and compute and
storage charges from cloud providers. With this
information, the CDO can determine how best
to optimize costs, such as routing requests by
priority level or moving certain data to the cloud
to cut down on networking costs.

The value of these metrics is only as great as the
degree to which CDOs act on them. CDOs will


need to establish data-performance metrics that
can be reviewed in near real time and protocols to
make rapid decisions. Effective data governance
programs should remain in place but be extended
to incorporate generative AI–related decisions.

Data cannot be an afterthought in generative AI.
Rather, it is the core fuel that powers the ability
of a business to capture value from generative
AI. But businesses that want that value cannot
afford CDOs who merely manage data; they need
CDOs who understand how to use data to lead
the business.


Joe Caserta is a partner in McKinsey’s New York office, where Kayvaun Rowshankish is a senior partner; Holger Harreis is
a senior partner in the Düsseldorf office, where Asin Tavakoli is a partner; and Nikhil Srinidhi is an associate partner in the
Berlin office.

The authors wish to thank Sven Blumberg, Stephanie Brauckmann, Carlo Giovine, Jonas Heite, Vishnu Kamalnath, Simon
Malberg, Rong Parnas, Bruce Philp, Adi Pradhan, Alex Singla, Saravanakumar Subramaniam, Alexander Sukharevsky, and
Kevin-Morris Wigand for their contributions to this article.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

##### Technology’s generational moment with generative AI: A CIO and CTO guide

CIOs and CTOs can take nine actions to reimagine business and technology
with generative AI.

_This article is a collaborative effort by Aamer Baig, Sven Blumberg, Eva Li, Douglas Merrill, Adi Pradhan,_
_Megha Sinha, Alexander Sukharevsky, and Stephen Xu, representing views from McKinsey Digital._

© Getty Images


-----

Hardly a day goes by without some new
business-busting development related to
generative AI surfacing in the media. The
excitement is well deserved—McKinsey research
estimates that generative AI could add the
equivalent of $2.6 trillion to $4.4 trillion of value
annually.¹

CIOs and chief technology officers (CTOs) have a
critical role in capturing that value, but it’s worth
remembering we’ve seen this movie before. New
technologies emerged—the internet, mobile,
social media—that set off a melee of experiments
and pilots, though significant business value
often proved harder to come by. Many of the
lessons learned from those developments still
apply, especially when it comes to getting past
the pilot stage to reach scale. For the CIO and
CTO, the generative AI boom presents a unique
opportunity to apply those lessons to guide the
C-suite in turning the promise of generative AI
into sustainable value for the business.

Through conversations with dozens of tech
leaders and an analysis of generative AI initiatives
at more than 50 companies (including our own),
we have identified nine actions all technology
leaders can take to create value, orchestrate
technology and data, scale solutions, and
manage risk for generative AI:

1. Move quickly to determine the company’s
posture for the adoption of generative AI,
and develop practical communications to, and
appropriate access for, employees.

2. Reimagine the business and identify use
cases that build value through improved
productivity, growth, and new business
models. Develop a “financial AI” (FinAI)
capability that can estimate the true costs
and returns of generative AI.


3. Reimagine the technology function,
and focus on quickly building generative
AI capabilities in software development,
accelerating technical debt reduction, and
dramatically reducing manual effort in IT
operations.

4. Take advantage of existing services or
adapt open-source generative AI models to
develop proprietary capabilities (building and
operating your own generative AI models can
cost tens to hundreds of millions of dollars, at
least in the near term).

5. Upgrade your enterprise technology
architecture to integrate and manage
generative AI models and orchestrate how
they operate with each other and existing
AI and machine learning (ML) models,
applications, and data sources.

6. Develop a data architecture to enable
access  to quality data by processing both
structured and unstructured data sources.

7. Create a centralized, cross-functional
generative AI platform team to provide
approved models to product and application
teams on demand.

8. Invest in upskilling key roles—software
developers, data engineers, MLOps
engineers, and security experts—as well as
the broader nontech workforce. But you need
to tailor the training programs by roles and
proficiency levels due to the varying impact
of generative AI.

9. Evaluate the new risk landscape and
establish ongoing mitigation practices to
address models, data, and policies.


1 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.


-----

**1. Determine the company’s posture for**
**the adoption of generative AI**
As use of generative AI becomes increasingly
widespread, we have seen CIOs and CTOs respond
by blocking employee access to publicly available
applications to limit risk. In doing so, these
companies risk missing out on opportunities for
innovation, with some employees even perceiving
these moves as limiting their ability to build
important new skills.

Instead, CIOs and CTOs should work with risk
leaders to balance the real need for risk mitigation
with the importance of building generative AI
skills in the business. This requires establishing
the company’s posture regarding generative AI
by building consensus around the levels of risk
with which the business is comfortable and how
generative AI fits into the business’s overall strategy.
This step allows the business to quickly determine
company-wide policies and guidelines.

Once policies are clearly defined, leaders should
communicate them to the business, with the CIO
and CTO providing the organization with appropriate
access and user-friendly guidelines. Some
companies have rolled out firmwide communications
about generative AI, provided broad access to
generative AI for specific user groups, created popups that warn users any time they input internal
data into a model, and built a guidelines page that
appears each time users access a publicly available
generative AI service.

**2. Identify use cases that build value**
**through improved productivity,**
**growth, and new business models**
CIOs and CTOs should be the antidote to the “death
by use case” frenzy that we already see in many
companies. They can be most helpful by working
with the CEO, CFO, and other business leaders
to think through how generative AI challenges


existing business models, opens doors to new
ones, and creates new sources of value. With a
deep understanding of the technical possibilities,
the CIO and CTO should identify the most valuable
opportunities and issues across the company that
can benefit from generative AI—and those that can’t.
In some cases, generative AI is not the best option.

McKinsey research, for example, shows generative
AI can lift productivity for certain marketing use
cases (for example, by analyzing unstructured and
abstract data for customer preference) by roughly
10 percent and customer support (for example,
through intelligent bots) by up to 40 percent.²
The CIO and CTO can be particularly helpful in
developing a perspective on how best to cluster use
cases either by domain (such as customer journey or
business process) or use case type (such as creative
content creation or virtual agents) so that generative
AI will have the most value. Identifying opportunities
won’t be the most strategic task—there are many
generative AI use cases out there—but, given initial
limitations of talent and capabilities, the CIO and
CTO will need to provide feasibility and resource
estimates to help the business sequence generative
AI priorities.

Providing this level of counsel requires tech leaders
to work with the business to develop a FinAI
capability to estimate the true costs and returns on
generative AI initiatives. Cost calculations can be
particularly complex because the unit economics
must account for multiple model and vendor costs,
model interactions (where a query might require
input from multiple models, each with its own fee),
ongoing usage fees, and human oversight costs.

**3. Reimagine the technology function**
Generative AI has the potential to completely
remake how the tech function works. CIOs and
CTOs need to make a comprehensive review of the
potential impact of generative AI on all areas of tech,


-----

but it’s important to take action quickly to build
experience and expertise. There are three areas
where they can focus their initial energies:

— Software development: McKinsey research
shows generative AI coding support can help
software engineers develop code 35 to 45
percent faster, refactor code 20 to 30 percent
faster, and perform code documentation 45
to 50 percent faster.³ Generative AI can also
automate the testing process and simulate
edge cases, allowing teams to develop
more-resilient software prior to release, and
accelerate the onboarding of new developers
(for example, by asking generative AI questions
about a code base). Capturing these benefits
will require extensive training (see more in
action 8) and automation of integration and
deployment pipelines through DevSecOps
practices to manage the surge in code volume.

— Technical debt: Technical debt can account for
20 to 40 percent of technology budgets and
significantly slow the pace of development.⁴
CIOs and CTOs should review their tech-debt
balance sheets to determine how generative
AI capabilities such as code refactoring,
code translation, and automated test-case
generation can accelerate the reduction of
technical debt.

— IT operations (ITOps): CIOs and CTOs will
need to review their ITOps productivity efforts
to determine how generative AI can accelerate
processes. Generative AI’s capabilities are
particularly helpful in automating such tasks
as password resets, status requests, or
basic diagnostics through self-serve agents;
accelerating triage and resolution through
improved routing; surfacing useful context,
such as topic or priority, and generating
suggested responses; improving observability
through analysis of vast streams of logs to
identify events that truly require attention; and


developing documentation, such as standard
operating procedures, incident postmortems,
or performance reports.

**4. Take advantage of existing services**
**or adapt open-source generative AI**
**models**
A variation of the classic “rent, buy, or build”
decision exists when it comes to strategies for
developing generative AI capabilities. The basic
rule holds true: a company should invest in a
generative AI capability where it can create a
proprietary advantage for the business and access
existing services for those that are more like
commodities.

The CIO and CTO can think through the implications
of these options as three archetypes:

— Taker—uses publicly available models through
a chat interface or an API, with little or no
customization. Good examples include offthe-shelf solutions to generate code (such
as GitHub Copilot) or to assist designers with
image generation and editing (such as Adobe
Firefly). This is the simplest archetype in
terms of both engineering and infrastructure
needs and is generally the fastest to get up
and running. These models are essentially
commodities that rely on feeding data in the
form of prompts to the public model.

— Shaper—integrates models with internal data
and systems to generate more customized
results. One example is a model that supports
sales deals by connecting generative AI
tools to customer relationship management
(CRM) and financial systems to incorporate
customers’ prior sales and engagement
history. Another is fine-tuning the model with
internal company documents and chat history
to act as an assistant to a customer support
agent. For companies that are looking to


2 Ibid.
3 Begum Karaci Deniz, Martin Harrysson, Alharith Hussin, and Shivam Srivastava, “Unleashing developer productivity with generative AI,”
McKinsey, June 27, 2023.
4 Vishal Dalal, Krish Krishnakanthan, Björn Münstermann, and Rob Patenge, “Tech debt: Reclaiming tech equity,” McKinsey, October 6, 2020.


-----

**5. Upgrade your enterprise technology**
**architecture to integrate and manage**
**generative AI models**
Organizations will use many generative AI models
of varying size, complexity, and capability. To
generate value, these models need to be able to
work both together and with the business’s existing
systems or applications. For this reason, building a
separate tech stack for generative AI creates more
complexities than it solves. As an example, we can
look at a consumer querying customer service at a
travel company to resolve a booking issue (Exhibit
2). In interacting with the customer, the generative
AI model needs to access multiple applications and
data sources.

For the Taker archetype, this level of coordination
isn’t necessary. But for companies looking to
scale the advantages of generative AI as Shapers
or Makers, CIOs and CTOs need to upgrade their
technology architecture. The prime goal is to
integrate generative AI models into internal systems
and enterprise applications and to build pipelines to
various data sources. Ultimately, it’s the maturity of
the business’s enterprise technology architecture
that allows it to integrate and scale its generative AI
capabilities.

Recent advances in integration and orchestration
frameworks, such as LangChain and LlamaIndex,
have significantly reduced the effort required to
connect different generative AI models with other
applications and data sources. Several integration
patterns are also emerging, including those that
enable models to call APIs when responding to
a user query—GPT-4, for example, can invoke
functions—and provide contextual data from an
external dataset as part of a user query, a technique
known as retrieval augmented generation. Tech
leaders will need to define reference architectures
and standard integration patterns for their
organization (such as standard API formats and
parameters that identify the user and the model
invoking the API).


scale generative AI capabilities, develop more
proprietary capabilities, or meet higher security
or compliance needs, the Shaper archetype is
appropriate.

There are two common approaches for
integrating data with generative AI models in
this archetype. One is to “bring the model to
the data,” where the model is hosted on the
organization’s infrastructure, either on-premises
or in the cloud environment. Cohere, for example,
deploys foundation models on clients’ cloud
infrastructure, reducing the need for data
transfers. The other approach is to “bring
data to the model,” where an organization can
aggregate its data and deploy a copy of the large
model on cloud infrastructure. Both approaches
achieve the goal of providing access to the
foundation models, and choosing between them
will come down to the organization’s workload
footprint.

— Maker—builds a foundation model to address
a discrete business case. Building a foundation
model is expensive and complex, requiring
huge volumes of data, deep expertise, and
massive compute power. This option requires
a substantial one-off investment—tens or
even hundreds of millions of dollars—to build
the model and train it. The cost depends on
various factors, such as training infrastructure,
model architecture choice, number of model
parameters, data size, and expert resources.

Each archetype has its own costs that tech
leaders will need to consider (Exhibit 1). While new
developments, such as efficient model training
approaches and lower graphics processing unit
(GPU) compute costs over time, are driving costs
down, the inherent complexity of the Maker
archetype means that few organizations will adopt
it in the short term. Instead, most will turn to some
combination of Taker, to quickly access a commodity
service, and Shaper, to build a proprietary capability
on top of foundation models.


-----

Exhibit 1
**Each archetype has its own costs.**

|Archetype|Example use cases|Estimated total cost of ownership|
|---|---|---|
|Taker|— Off-the-shelf coding assistant for software developers — General-purpose customer service chatbot with prompt engineering only and text chat only|~ $0.5 million to $2.0 million, onetime — Off-the-shelf coding assistant: ~$0.5 million for integration. Costs include a team of 6 working for 3 to 4 months. — General-purpose customer service chatbot: ~$2.0 million for building plug-in layer on top of third- party model API. Costs include a team of 8 working for 9 months. ~ $0.5 million, recurring annually — Model inference: • Off-the-shelf coding assistant: ~$0.2 million annually per 1,000 daily users • General-purpose customer service chatbot: ~$0.2 million annually, assuming 1,000 customer chats per day and 10,000 tokens per chat — Plug-in-layer maintenance: up to ~$0.2 million annually, assuming 10% of development cost.|
|Shaper|— Customer service chatbot fine-tuned with sector-specific knowledge and chat history|~ $2.0 million to $10.0 million, onetime unless model is fine-tuned further — Data and model pipeline building: ~$0.5 million. Costs include 5 to 6 machine learning engineers and data engineers working for 16 to 20 weeks to collect and label data and perform data ETL.¹ — Model fine-tuning²: ~$0.1 million to $6.0 million per training run³ • Lower end: costs include compute and 2 data scientists working for 2 months • Upper end: compute based on public closed-source model fine-tuning cost — Plug-in-layer building: ~$1.0 million to $3.0 million. Costs include a team of 6 to 8 working for 6 to 12 months. ~ 0.5 million to $1.0 million, recurring annually — Model inference: up to ~$0.5 million recurring annually. Assume 1,000 chats daily with both audio and texts. — Model maintenance: ~$0.5 million. Assume $100,000 to $250,000 annually for MLOps platform⁴ and 1 machine learning engineer spending 50% to 100% of their time monitoring model performance. — Plug-in-layer maintenance: up to ~$0.3 million recurring annually, assuming 10% of development cost.|
|Maker|— Foundation model trained for assisting in patient diagnosis|~ $5.0 million to $200.0 million, onetime unless model is fine-tuned or retrained — Model development: ~$0.5 million. Costs include 4 data scientists spending 3 to 4 months on model design, development, and evaluation leveraging existing research. — Data and model pipeline: ~$0.5 million to $1.0 million. Costs include 6 to 8 machine learning engineers and data engineers working for ~12 weeks to collect data and perform data ETL.¹ — Model training⁵: ~$4.0 million to $200.0 million per training run.³ Costs include compute and labor cost of 4 to 6 data scientists working for 3 to 6 months. — Plug-in-layer building: ~$1.0 million to $3.0 million. Costs include a team of 6 to 8 working 6 to 12 months. ~ $1.0 million to $5.0 million, recurring annually — Model inference: ~$0.1 million to $1.0 million annually per 1,000 users. Assume each physician sees 20 to 25 patients per day and patient speaks for 6 to 25 minutes per visit. — Model maintenance: ~$1.0 million to $4.0 million recurring annually. Assume $250,000 annually for MLOps platform⁴ and 3 to 5 machine learning engineers to monitor model performance. — Plug-in-layer maintenance: up to ~$0.3 million recurring annually, assuming 10% of development cost.|



Note: Through engineering optimizations, the economics of generative AI are evolving rapidly, and these are high-level estimates based on total cost of ownership (resources,
model training, etc) as of mid-2023.

1 Extract, transform, and load.
² Model is fine-tuned on dataset consisting of ~100,000 pages of sector-specific documents and 5 years of chat history from ~1,000 customer representatives, which is ~48
billion tokens. Lower end cost consists of 1% parameters retrained on open-source models (eg, LLaMA) and upper end on closed-source models. Chatbot can be accessed
via both text and audio.
³ Model is optimized after each training run based on use of hyperparameters, dataset, and model architecture. Model may be refreshed periodically when needed (eg, with
fresh data).
⁴ Gilad Shaham, “Build or buy your MLOps platform: Main considerations,” LinkedIn, November 3, 2021.
5 Model is trained on 65 billion to 1 trillion parameters and dataset of 1 2 to 2 4 trillion tokens The tool can be accessed via both text and audio


-----

Exhibit 2
**Generative AI is integrated at key touchpoints to enable a tailoredGenerative AI is integrated at key touchpoints to enable a tailored**
**customer journey.customer journey.**

Illustrative customer journey using travel agent bot

|r r|C li reviews|
|---|---|
|||
|Disagrees||


|r requests Customer completes book-|Col2|Col3|
|---|---|---|
|t ing change and drops of|||
|Chatbot Agent Agent pings picks up inputs cus- case new solu- tomer and tion for support provides review/ new feedback solution to model|||
|Model instructs customer o support system to assign agent|||
||||
||||


Cus- Customer logs in and requests Customer reviews Customer requests Customer completes booktomer to change booking options live agent ing change and drops of

_Disagrees_

Inter- _Chatbot_ _Chatbot_ _Chatbot_ _Chatbot_ _Agent_ _Agent_
action activated communi- re- pings picks up inputs

cates sponds cus- case new solumessage tomer and tion for

_Selects_

and support provides review/

_option_

options new feedback

solution to model

Genera- Model Model Model Model Model
tive AI receives checks explains instructs instructs
model user booking issue and booking customer

request policy and gives system to support
and pulls sees cus- alternate complete system to
user info in tomer can- options task assign
prompt not make agent

change

Back- Log-in authentif- Booking Workfow Workfow
end apps cation, model/cus- modifcation management management for

tomer info access policy for booking live agent
authorization management assignment

Data
source

Customer ID data Customer history Policy data Booking system Agent assignment
data data data


Infra
and
compute

Cloud/on-premises infrastructure and compute


McKinsey & Company


API calls


-----

There are five key elements that need to be
incorporated into the technology architecture to
integrate generative AI effectively (Exhibit 3):

— Context management and caching to
provide models with relevant information from
enterprise data sources. Access to relevant
data at the right time is what allows the model to
understand the context and produce compelling
outputs. Caching stores results to frequently
asked questions to enable faster and cheaper
responses.

— Policy management to ensure appropriate
access to enterprise data assets. This control
ensures that HR’s generative AI models that
include employee compensation details, for
example, cannot be accessed by the rest of the
organization.

— Model hub, which contains trained and
approved models that can be provisioned on
demand and acts as a repository for model
checkpoints, weights, and parameters.

— Prompt library, which contains optimized
instructions for the generative AI models,
including prompt versioning as models are
updated.

— MLOps platform, including upgraded MLOps
capabilities, to account for the complexity of
generative AI models. MLOps pipelines, for
example, will need to include instrumentation
to measure task-specific performance, such as
measuring a model’s ability to retrieve the right
knowledge.

In evolving the architecture, CIOs and CTOs will
need to navigate a rapidly growing ecosystem of
generative AI providers and tooling. Cloud providers
provide extensive access to at-scale hardware
and foundation models, as well as a proliferating
set of services. MLOps and model hub providers,
meanwhile, offer the tools, technologies, and
practices to adapt a foundation model and deploy
it into production, while other companies provide
applications directly accessed by users built on


top of foundation models to perform specific
tasks. CIOs and CTOs will need to assess how
these various capabilities are assembled and
integrated to deploy and operate generative AI
models.

**6. Develop a data architecture to**
**enable access to quality data**
The ability of a business to generate and
scale value, including cost reductions and
improved data and knowledge protections, from
generative AI models will depend on how well it
takes advantage of its own data. Creating that
advantage relies on a data architecture that
connects generative AI models to internal data
sources, which provide context or help fine-tune
the models to create more relevant outputs.

In this context, CIOs, CTOs, and chief data
officers need to work closely together to do the
following:

— Categorize and organize data so it can be
used by generative AI models. Tech leaders
will need to develop a comprehensive
data architecture that encompasses both
structured and unstructured data sources.
This requires putting in place standards and
guidelines to optimize data for generative AI
use—for example, by augmenting training
data with synthetic samples to improve
diversity and size; converting media types
into standardized data formats; adding
metadata to improve traceability and data
quality; and updating data.

— Ensure existing infrastructure or cloud
services can support the storage and
handling of the vast volumes of data needed
for generative AI applications.

— Prioritize the development of data pipelines
to connect generative AI models to relevant
data sources that provide “contextual
understanding.” Emerging approaches
include the use of vector databases to store
and retrieve embeddings (specially formatted


-----

Exhibit 3
**The tech stack for generative AI is emerging.**

The tech stack for generative AI is emerging.


Illustrative generative AI tech stack

Apps Models Data

Tooling Infrastructure

Apps-as-a- Data sources Experience layer Policy
service with Embeddings, DTC² or B2B applications (eg, Jasper) management
embedded unstructured

Role-based

foundation data,

access

models analytical

API gateway control and

End-user- data, trans- contentfacing actional data based policies
applications to secure
and founda- Context management and caching enterprise
tion models data assets

User and task context retrieved from enterprise data

accessed

sources to prompt generative AI models, cache for

through a

common requests

browser
interface as
SaaS¹ (eg,
Midjourney)

Data Model hub Prompt
platforms library

Platforms that allow users to share

Vector models and datasets (eg, Hugging Face)
databases,
data
warehouse,
data lake Closed-source Open-/closed-source

foundation foundation models
models Trained model that is made

API-based, pre- accessible (eg, BLOOM)
trained models
(eg, GPT-4)

MLOps platform

Existing enterprise platforms
(eg, ERP,³ CRM⁴)

Cloud or on-premises infrastructure QA and
and compute hardware observability

QA model
outputs (eg,
checks for bias)


1Software as a service.
2Direct to consumer.
3Enterprise resource planning.
4Customer relationship management.

McKinsey & Company


Users


-----

knowledge) as input for generative AI models as
well as in-context learning approaches, such as
“few shot prompting,” where models are provided
with examples of good answers.

**7. Create a centralized, cross-functional**
**generative AI platform team**
Most tech organizations are on a journey to a
product and platform operating model. CIOs and
CTOs need to integrate generative AI capabilities
into this operating model to build on the existing
infrastructure and help to rapidly scale adoption
of generative AI. The first step is setting up a
generative AI platform team whose core focus is
developing and maintaining a platform service
where approved generative AI models can be
provisioned on demand for use by product and
application teams. The platform team also defines
protocols for how generative AI models integrate
with internal systems, enterprise applications,
and tools, and also develops and implements
standardized approaches to manage risk, such as
responsible AI frameworks.

CIOs and CTOs need to ensure that the platform
team is staffed with people who have the right
skills. This team requires a senior technical leader
who acts as the general manager. Key roles include
software engineers to integrate generative AI
models into existing systems, applications, and
tools; data engineers to build pipelines that
connect models to various systems of record and
data sources; data scientists to select models and
engineer prompts; MLOps engineers to manage
deployment and monitoring of multiple models and
model versions; ML engineers to fine-tune models
with new data sources; and risk experts to manage
security issues such as data leakage, access
controls, output accuracy, and bias. The exact
composition of the platform team will depend on
the use cases being served across the enterprise. In
some instances, such as creating a customer-facing
chatbot, strong product management and user
experience (UX) resources will be required.


Realistically, the platform team will need to work
initially on a narrow set of priority use cases,
gradually expanding the scope of their work as they
build reusable capabilities and learn what works
best. Technology leaders should work closely with
business leads to evaluate which business cases to
fund and support.

**8. Tailor upskilling programs by roles**
**and proficiency levels**
Generative AI has the potential to massively
lift employees’ productivity and augment their
capabilities. But the benefits are unevenly
distributed depending on roles and skill levels,
requiring leaders to rethink how to build the actual
skills people need.

Our latest empirical research using the generative
AI tool GitHub Copilot, for example, helped software
engineers write code 35 to 45 percent faster.⁵ The
benefits, however, varied. Highly skilled developers
saw gains of up to 50 to 80 percent, while junior
developers experienced a 7 to 10 percent decline in
speed. That’s because the output of the generative
AI tools requires engineers to critique, validate,
and improve the code, which inexperienced
software engineers struggle to do. Conversely, in
less technical roles, such as customer service,
generative AI helps low-skill workers significantly,
with productivity increasing by 14 percent and staff
turnover dropping as well, according to one study.⁶

These disparities underscore the need for
technology leaders, working with the chief human
resources officer (CHRO), to rethink their talent
management strategy to build the workforce of the
future. Hiring a core set of top generative AI talent
will be important, and, given the increasing scarcity
and strategic importance of that talent, tech leaders
should put in place retention mechanisms, such
as competitive salaries and opportunities to be
involved in important strategic work for the business.

Tech leaders, however, cannot stop at hiring.
Because nearly every existing role will be affected


5 “Unleashing developer productivity with generative AI,” McKinsey, June 27, 2023.


-----

**9. Evaluate the new risk landscape**
**and establish ongoing mitigation**
**practices**
Generative AI presents a fresh set of ethical
questions and risks, including “hallucinations,”
whereby the generative AI model presents
an incorrect response based on the highestprobability response; the accidental release of
confidential personally identifiable information;
inherent bias in the large datasets the models
use; and high degrees of uncertainty related to
intellectual property (IP). CIOs and CTOs will need
to become fluent in ethics, humanitarian, and
compliance issues to adhere not just to the letter
of the law (which will vary by country) but also to
the spirit of responsibly managing their business’s
reputation.

Addressing this new landscape requires a
significant review of cyber practices and updating
the software development process to evaluate
risk and identify mitigation actions before model
development begins, which will both reduce
issues and ensure the process doesn’t slow down.
Proven risk-mitigation actions for hallucinations
can include adjusting the level of creativity
(known as the “temperature”) of a model when it
generates responses; augmenting the model with
relevant internal data to provide more context;
using libraries that impose guardrails on what
can be generated; using “moderation” models
to check outputs; and adding clear disclaimers.
Early generative AI use cases should focus on
areas where the cost of error is low, to allow the
organization to work through inevitable setbacks
and incorporate learnings.

To protect data privacy, it will be critical to
establish and enforce sensitive data tagging
protocols, set up data access controls in different
domains (such as HR compensation data), add
extra protection when data is used externally,


by generative AI, a crucial focus should be on
upskilling people based on a clear view of what
skills are needed by role, proficiency level, and
business goals. Let’s look at software developers
as an example. Training for novices needs to
emphasize accelerating their path to become top
code reviewers in addition to code generators.
Similar to the difference between writing and
editing, code review requires a different skill set.
Software engineers will need to understand what
good code looks like; review the code created
by generative AI for functionality, complexity,
quality, and readability; and scan for vulnerabilities
while ensuring they do not themselves introduce
quality or security issues in the code. Furthermore,
software developers will need to learn to think
differently when it comes to coding, by better
understanding user intent so they can create
prompts and define contextual data that help
generative AI tools provide better answers.

Beyond training up tech talent, the CIO and CTO
can play an important role in building generative
AI skills among nontech talent as well. Besides
understanding how to use generative AI tools for
such basic tasks as email generation and task
management, people across the business will
need to become comfortable using an array of
capabilities to improve performance and outputs.
The CIO and CTO can help adapt academy models
to provide this training and corresponding
certifications.

The decreasing value of inexperienced engineers
should accelerate the move away from a classic
talent pyramid, where the greatest number of
people are at a junior level, to a structure more
like a diamond, where the bulk of the technical
workforce is made up of experienced people.
Practically speaking, that will mean building the
skills of junior employees as quickly as possible
while reducing roles dedicated to low-complexity
manual tasks (such as writing unit tests).


6 Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, Generative AI at work, National Bureau of Economic Research (NBER) working paper,
number 31161, April 2023.


-----

and include privacy safeguards. For example, to
mitigate access control risk, some organizations
have set up a policy-management layer that
restricts access by role once a prompt is given to
the model. To mitigate risk to intellectual property,
CIOs and CTOs should insist that providers
of foundation models maintain transparency
regarding the IP (data sources, licensing, and
ownership rights) of the datasets used.


Generative AI is poised to be one of the fastestgrowing technology categories we’ve ever seen.
Tech leaders cannot afford unnecessary delays
in defining and shaping a generative AI strategy.
While the space will continue to evolve rapidly,
these nine actions can help CIOs and CTOs
responsibly and effectively harness the power of
generative AI at scale.


Aamer Baig is a senior partner in McKinsey’s Chicago office; Sven Blumberg is a senior partner in the Düsseldorf office; Eva
Li is a consultant in the Bay Area office, where Megha Sinha is a partner; Douglas Merrill is a partner in the Southern California
office; Adi Pradhan and Stephen Xu are associate partners in the Toronto office; and Alexander Sukharevsky is a senior
partner in the London office.

The authors wish to thank Stephanie Brauckmann, Anusha Dhasarathy, Martin Harrysson, Klemens Hjartar, Alharith Hussin,
Naufal Khan, Sam Nie, Chandrasekhar Panda, Henning Soller, Nikhil Srinidhi, Asin Tavakoli, Niels Van der Wildt, and Anna
Wiesinger for their contributions to this article.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

Risk & Resilience Practice
**As gen AI advances,**
**regulators—and risk**
**functions—rush to keep pace**

AI and its supercharged breakthrough, generative AI, are all about rapid
advancements, and rule makers are under pressure to keep up.

_This article is a collaborative effort by Andreas Kremer, Angela Luget, Daniel Mikkelsen, Henning Soller, Malin_
_Strandell-Jansson, and Sheila Zingg, representing views from McKinsey’s Risk & Resilience Practice._

© Getty Images


-----

The rapid advancement of generative AI (gen
AI) has regulators around the world racing to
understand, control, and guarantee the safety of
the technology—all while preserving its potential
benefits. Across industries, gen AI adoption
has presented a new challenge for risk and
compliance functions: how to balance use of this
new technology amid an evolving—and uneven—
regulatory framework.

As governments and regulators try to define what
such a control environment should look like, the
developing approaches are fragmented and often
misaligned, making it difficult for organizations to
navigate and causing substantial uncertainty.

In this article, we explain the risks of AI and gen
AI and why the technology has drawn regulatory
scrutiny. We also offer a strategic road map to help
risk functions navigate the uneven and changing
rule-making landscape—which is focused not only
on gen AI but all artificial intelligence.

**Why does gen AI need regulation?**
AI’s breakthrough advancement, gen AI, has
quickly captured the interest of the public, with
ChatGPT becoming one of the fastest-growing
platforms ever, reaching one million users in just
five days. The acceleration comes as no surprise
given the wide range of gen AI use cases, which
promise increased productivity, expedited access
to knowledge, and an expected total economic
impact of $2.6 trillion to $4.4 trillion annually.¹

There is, however, an economic incentive to
getting AI and gen AI adoption right. Companies
developing these systems may face consequences
if the platforms they develop are not sufficiently
polished. And a misstep can be costly. Major gen
AI companies, for example, have lost significant
market value when their platforms were found
hallucinating (when AI generates false or illogical
information).

The proliferation of gen AI has increased the
visibility of risks. Key gen AI concerns include


how the technology’s models and systems are
developed and how the technology is used.

Generally, there are concerns about a potential
lack of transparency in the functioning of gen AI
systems, the data used to train them, issues of
bias and fairness, potential intellectual property
infringements, possible privacy violations, thirdparty risk, as well as security concerns.

Add disinformation to these concerns, such as
erroneous or manipulated output and harmful or
malicious content, and it is no wonder regulators
are seeking to mitigate potential harms. Regulators
seek to establish legal certainty for companies
engaged in the development or use of gen AI.
Meanwhile, rule makers want to encourage
innovation without fear of unknown repercussions.

The goal is to establish harmonized international
regulatory standards that would stimulate
international trade and data transfers. In pursuit
of this goal, a consensus has been reached: the
gen AI development community has been at the
forefront of advocating for some regulatory control
over the technology’s development as soon as
possible. The question at hand is not whether to
proceed with regulations, but rather how.

**The current international regulatory**
**landscape for AI**
While no country has passed comprehensive AI
or gen AI regulation to date, leading legislative
efforts include those in Brazil, China, the European
Union, Singapore, South Korea, and the United
States. The approaches taken by the different
countries vary from broad AI regulation supported
by existing data protection and cybersecurity
regulations (the European Union and South Korea)
to sector-specific laws (the United States) and
more general principles or guidelines-based
approaches (Brazil, Singapore, and the United
States). Each approach has its own benefits and
drawbacks, and some markets will move from
principles-based guidelines to strict legislation
over time (Exhibit 1).


1 “The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.


-----

Web 2023
GenAIRegulations
Exhibit 1 of 2

Exhibit 1
**Regulations related to AI governance vary around the world.Regulations related to AI governance vary around the world.**

As of November 2023, nonexhaustive

Type of policy: General AI legislation proposed Example countries without general
Nonbinding principles (eg, OECD) or being fnalized AI legislation


Japan
Singapore
United Arab Emirates
United Kingdom
United States
Other OECD member countries

Source: OECD; McKinsey analysis

McKinsey & Company


Brazil
Canada
China
South Korea
European Union


Australia
India
New Zealand
Saudi Arabia

failures to meet these criteria, and they should
be resilient against attempts to manipulate the
system by malicious third parties.

— Diversity, nondiscrimination, and fairness.
Another goal for regulators is to ensure that AI
systems are free of bias and that the output
does not result in discrimination or unfair
treatment of people.

— Privacy and data governance. Regulators want
to see development and usage of AI systems
that follow existing privacy and data protection
rules while processing data that meets high
standards in quality and integrity.

— Social and environmental well-being. There is a
strong desire to ensure that all AI is sustainable,
environmentally friendly (for instance, in its
energy use), and beneficial to all people, with
ongoing monitoring and assessing of the
long-term effects on individuals, society, and

democracy.

Despite some commonality in the guiding principles
of AI, the implementation and exact wording vary by
regulator and region. Many rules are still new and,
thus, prone to frequent updates (Exhibit 2). This
makes it challenging for organizations to navigate
regulations while planning long-term AI strategies.


While the approaches vary, common themes in the
regulatory landscape have emerged globally:

— Transparency. Regulators are seeking traceability
and clarity of AI output. Their goal is to ensure
that users are informed when they engage
with any AI system and to provide them with
information about their rights and about the
capabilities and limitations of the system.

— Human agency and oversight. Ideally, AI
systems should be developed and used as tools
that serve people, uphold human dignity and
personal autonomy, and function in a way that
can be appropriately controlled and overseen by
humans.

— Accountability. Regulators want to see
mechanisms that ensure awareness of
responsibilities, accountability, and potential
redress regarding AI systems. In practice,
they are seeking top management buy-in,
organization-wide education, and awareness of
individual responsibility.

— Technical robustness and safety. Rule makers are
seeking to minimize unintended and unexpected
harm by ensuring that AI systems are robust,
meaning they operate as expected, remain stable,
and can rectify user errors. They should have
fallback solutions and remediation to address any


-----

Web 2023
GenAIRegulations

Exhibit 2 Exhibit 2 of 2
**AI governance–related policy and regulatory efforts are under way globally.**

AI governance–related policy and regulatory eforts are under way globally.

Examples by type of policy or efort and when proposed; nonexhaustive

Nonbinding principles (eg, OECD) General AI legislation proposed Vertical-specifc AI regulations







21


25



22 23 24


|Col1|4|
|---|---|
|3||
|||

|9|Col2|
|---|---|
|||

|14|15|
|---|---|
|||


17

16 18


|6 18 20|Col2|19 2|0 2|Col5|2 2|
|---|---|---|---|---|---|
|||||||
|||||||


1

2

2019

7

5 6 8

2020 2021

10

11 12 13

2022

2019 and earlier 2020 2021 2022 2023

|24|26|Col3|
|---|---|---|
||||
||||



-  1. Sept 2017
South Korea Ethical
Guidelines for
Intelligent Information
Technology

-  2. Jan 2019
Singapore Model
AI Governance
Framework, frst edition

-  3. Apr 2019
EU Ethics Guidelines
for Trustworthy AI

-  4. May 2019
OECD AI Principles

-  5. Sept 2019
Bill establishing the
principles for the use
of AI in Brazil



-  6. Jan 2020
Singapore Model
AI Governance
Framework, second
edition

-  7. Feb 2020
Bill establishing the
fundamental principles
and guidelines for the
development and
application of AI in
Brazil

-  8. June 2020
South Korea
Framework Act
on Intelligent
Informatization



-  9. Mar 2021
Bill providing for the
ethical framework
and guidelines that
underlie the
development and
use of AI in Brazil

-  10. Apr 2021
Proposed EU AI Act
(expires Q1 2024)

South Korea

- 11. June 2021

Enforcement decree
on Framework Act
on Intelligent
Informatization


China issues provisions

- 12. Mar 2022

on Internet Information
Service Algorithm
Recommendations and
Administration of Deep
Synthesis of Internet
Information Services

Canada’s proposed

- 13. June 2022

Artifcial Intelligence
and Data Act (planned
2025)

EU AI Liability

- 14. Sept 2022

Directive, a regime for
dealing with damages
caused by AI

-  15. Oct 2022
US Blueprint for an
AI Bill of Rights

-  16. Dec 2022
Senate approval of
the draft regulatory
framework on artifcial
intelligence in Brazil


Stable Difusion and

- 17. Jan 2023

Midjourney copyright
lawsuits in the US

-  18. Jan 2023
NIST AI risk
management
framework

South Korean

- 19. Feb 2023

Assembly proposed Act
on Promotion of AI
Industry and Framework for Establishing
Trustworthy AI

-  20. Mar 2023
ChatGPT temporarily
banned in Italy because
of privacy concerns

-  21. Mar–Apr 2023
Several data protection
regulators globally
looking into ChatGPT
data protection
practices, eg, Germany,
France, and Spain

-  22. Apr 2023
China released
Draft Administrative
Measures for
Generative Artifcial
Intelligence Services

-  23. May 2023
Proposal for legal
framework for
artifcial intelligence
in Brazil merging
previous proposals
from 2019–21

-  24. Oct 2023
US presidential
executive order on AI

-  25. Nov 2023
AI summit in UK

Political agreement

- 26. Dec 2023

on EU AI Act


Source: OECD; McKinsey analysis

McKinsey & Company


-----

**What does this mean for organizations?**
Organizations may be tempted to wait to see what
AI regulations emerge. But the time to act is now.
Organizations may face large legal, reputational,
organizational, and financial risks if they do not act
swiftly. Several markets, including Italy, have already
banned ChatGPT because of privacy concerns,
copyright infringement lawsuits brought by multiple
organizations and individuals, and defamation
lawsuits.

More speed bumps are likely. As the negative
effects of AI become more widely known and
publicized, public concerns increase. This, in turn,
has led to public distrust of the companies creating
or using AI.

A misstep at this stage could also be costly.
Organizations could face fines from legal
enforcement—of up to 7 percent of annual global
revenues, according to the AI regulation proposed
by the European Union, for example. Another threat
is financial loss from falloff in customer or investor
trust that could translate into a lower stock price,
loss of customers, or slower customer acquisition.
The incentive to move fast is heightened by the
fact that if the right governance and organizational
models for AI are not built early, remediation may
become necessary later due to regulatory changes,
data breaches, or cybersecurity incidents. Fixing
a system after the fact can be both expensive
and difficult to implement consistently across the
organization.

The exact future of legal obligations is still unclear
and may differ across geographies and depend


on the specific role AI will play within the value
chain. Still, there are some no-regret moves for
organizations, which can be implemented today
to get ahead of looming legal changes.

These preemptive actions can be grouped into
four key areas that stem from existing data
protection or privacy and cyber efforts, as they
share a great deal of common ground:

_Transparency. Create a taxonomy and inventory_
of models, classifying them in accordance with
regulation, and record all usage across the
organization in a central repository that is clear
to those inside and outside the organization.
Create detailed documentation of AI and gen
AI usage, both internally and externally, its
functioning, risks, and controls, and create clear
documentation on how a model was developed,
what risks it may have, and how it is intended to
be used.

_Governance. Implement a governance structure_
for AI and gen AI that ensures sufficient oversight,
authority, and accountability both within the
organization and with third parties and regulators.
This approach should include a definition of
all roles and responsibilities in AI and gen AI
management and the development of an incident
management plan to address any issues that may
arise from AI and gen AI use. The governance
structure should be robust enough to withstand
changes in personnel and time but also agile
enough to adapt to evolving technology, business
priorities, and regulatory requirements.


**Organizations are challenged with**
**navigating varied regulations while**
**planning their long-term AI strategies.**


-----

_Data, model, and technology management. AI_
and gen AI both require robust data, model, and
technology management:

— Data management. Data is the foundation of
all AI and gen AI models. The quality of the
data input also mirrors the final output of the
model. Proper and reliable data management
includes awareness of data sources, data
classification, data quality and lineage,
intellectual property, and privacy management.

— Model management. Organizations can
establish robust principles and guardrails
for AI and gen AI development and use them
to minimize the organization’s risks and
ensure that all AI and gen AI models uphold
fairness and bias controls, proper functioning,
transparency, clarity, and enablement of
human oversight. Train the entire organization
on the proper use and development of AI and
gen AI to ensure risks are minimized. Develop
the organization’s risk taxonomy and risk
framework to include the risks associated with
gen AI. Establish roles and responsibilities
in risk management and establish risk
assessments and controls, with proper testing
and monitoring mechanisms to monitor and
resolve AI and gen AI risks. Both data and
model management require agile and iterative
processes and should not be treated as simple
tick-the-box exercises at the beginning of
development projects.


— Cybersecurity and technology management.
Establish strong cybersecurity and technology,
including access control, firewalls, logs,
monitoring, et cetera, to ensure a secure
technology environment, where unauthorized
access or misuse is prevented and potential
incidents are identified early.

_Individual rights. Educate users: make them aware_
that they are interacting with an AI system, and
provide clear instructions for use. This should
include establishing a point of contact that
provides transparency and enables users to
exercise their rights, such as how to access data,
how models work, and how to opt out. Finally,
take a customer-centric approach to designing
and using AI, one that considers the ethical
implications of the data used and its potential
impact on customers. Since not everything legal is
necessarily ethical, it is important to prioritize the
ethical considerations of AI usage.

AI and gen AI will continue to have a significant
impact on many organizations, whether
they are providers of AI models or users of
AI systems. Despite the rapidly changing
regulatory landscape, which is not yet aligned
across geographies and sectors and may feel
unpredictable, there are tangible benefits for
organizations that improve how they provide and
use AI now.


Andreas Kremer is a partner in McKinsey’s Berlin office; Angela Luget is a partner in the London office, where Daniel
Mikkelsen is a senior partner; Henning Soller is a partner in the Franfurt office; Malin Strandell-Jansson is a senior
knowledge expert in the Stockholm office; and Sheila Zingg is a consultant in the Zurich office.

The authors wish to thank Rachel Lee, Chris Schmitz, and Angie Selzer for their contributions to this article.

This article was edited by David Weidner, a senior editor in the Bay Area office.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

### What the future holds


-----

**Six major gen AI trends that**
**will shape 2024’s agenda**

What every CEO needs to know and what it means for their gen AI agendas.

_by Sam Bourton, Ben Ellencweig, Carlo Giovine, and Stephen Xu_


© Getty Images


-----

It’s hard to believe that ChatGPT is only a year
old. The number of exciting new product launches
over the past 12 months has been astonishing—
and there’s no sign that things will slow down
anytime soon. In fact, quite the opposite. Earlier
in November, OpenAI hosted DevDay, where the
company announced extensive offerings across
B2C and B2B markets. Cohere has doubled
down on its knowledge search capabilities and
private deployments. And Amazon Web Services
launched PartyRock, its no-code gen AI appbuilding playground.

We believe that this past month’s activity is
setting the stage for what to expect in 2024
in the gen AI space. Here are six major trends
happening across the space.

1. Gen AI can see, hear, and talk. Multimodal
models spanning text, code, image, and audio
unlock new capabilities across both content
generation and comprehension.

2. Gen AI can interact with the world. Gen AI
models connect to data and IT systems to
read and write data.

3. Gen AI models are easier to control. End
users get more consistent outputs from
probabilistic models via new features (for
example, setting the seed).

4. Gen AI development is being democratized.
OpenAI announced a new product called
“GPTs,” which allow nearly anyone to build a
gen-AI-powered chatbot using low code/no
code interfaces.

5. Gen AI is a platform play. Entire
marketplaces of GPTs will be created. There,
users will be able to discover new applications
and publish their own.

6. Gen AI costs continue to decline. For one,
GPT-4 API costs declined two to three times
for the average enterprise customer.


While the technology’s possibilities continue to
grow, we believe there are four principles for CEOs
to consider as they drive their gen AI agendas. The
principles draw from our experiences building gen
AI applications with our clients throughout 2023,
as well as decades delivering digital and analytics
transformations (covered thoroughly in our book,
_Rewired)._

Be intentional. Set gen AI strategy top-down.
Gen AI is a gold rush. Everyone from shareholders to
employees to boards are scrambling to deploy the
latest and most powerful gen AI tooling, and many
large organizations have 150-plus gen AI use cases
on backlog. While we share their excitement (and
admire their ambition!), we have found that allowing
dozens of gen AI projects to spawn across an
organization puts at-scale value creation at risk.

With the recent developments in the gen AI
space, the Cambrian explosion of the use cases
and opportunities will only continue to split the
already divided attention of leadership teams.
C-suites must bring focus with a top-down gen AI
strategy, while constantly returning to the question
of how the technology can help create enduring
strategic distance between the organization and
its competitors. Here are some examples from first
movers:

— Retail banks are increasing customer retention
and offering conversion by deploying customerfacing chatbots and hyperpersonalization.

— Service operations are tackling ongoing labor
shortages by building workflow copilots to
improve productivity of existing labor forces to
resolve customer requests on time.

— IT services players are growing market share by
investing in software engineering productivity
tools and pricing contracts more competitively.

Smart organizations are taking a 2x2 approach:
identify two fast use cases to register quick wins on
the scoreboard and excite the organization, while


-----

strategies will continue to play a central role in any
gen AI strategy.

Meanwhile, some organizations are finding
compelling business cases to “build,” as well. These
players start by identifying use cases that create
strategic competitive advantages against their peers,
by compounding existing strengths in their domain
expertise, workflow integration, or regulatory knowhow. For example, deploying gen AI to accelerate
drug discovery has become standard in the
pharmaceutical industry. Additionally, organizations
are making investments in data and IT infrastructure
to enable their portfolio of gen AI use cases. For many
organizations, there has been little to no investment
in unstructured data governance. Now is the time.

Build products, not POCs.
With the new tooling available, a talented engineer
can build a proof of concept over a weekend. In some
cases, this might be sufficient to serve an enterprise
need (for example, a summarization chatbot).
However, for most use cases in a large enterprise
context, proofs of concepts are not sufficient. They do
not scale well into production and their performance
rapidly degrades without the appropriate engineering
and experimentation. At OpenAI’s Dev Day, engineers
showed how hard it is to turn a POC into a productiongrade product. At the start, a demo POC only
achieved 45 percent accuracy for a retrieval task.
After a few months and a dozen or so experiments (for
example, fine-tuning, reranking, metadata tagging,
data labeling, model self-assessment, risk guardrails),
the engineers achieved 98 percent accuracy.

This leads to two implications. First, organizations
cannot seek near-perfection on every use case. They
need to be selective about when it is worthwhile to
invest scarce engineering talent to develop highperformance gen AI applications. For some situations,
45 percent accuracy may be sufficient to deliver
business benefit. Second, organizations need to scale
their gen AI capabilities to meet their ambitions. Most
organizations have identified hundreds of gen AI use
cases. And so, organizations are turning to reusable
code components to accelerate development.
Dedicated engineers, often sitting in a center of
excellence (COE), codify best practices into these


working on two slower, more transformational
use cases that will change day-to-day business
operations.

Reimagine entire domains rather than isolated
use cases.
During 2023, most organizations began
experimenting with gen AI, building one-off
prototypes and buying off-the-shelf solutions.
Yet as these solutions are rolled out to end users,
organizations are struggling to capture value. For
example, some organizations that have invested
in Github Copilot have yet to figure out how the
value capture is passed back to the business.
Organizations need to reframe from one isolated use
case to the full software delivery life cycle. Scrum
teams need to commit to shipping more product
features. Or sales needs to offer more competitive
pricing to their customers and win more business. If
companies stop at just buying a new shiny tool, the
productivity gains will not translate to bottom-line
gains.

That often means reimagining entire workflows and
domains. This serves two purposes: 1) it creates a
more seamless end-user experience by avoiding
point solutions; and 2) organizations can more
easily track value against clear business outcomes.
For example, an insurer we have worked with is
reimagining its end-to-end claims process—from
first notice of loss to payment. For each step along
the way, the insurer has identified gen AI, digital, and
analytics opportunities, while never losing sight of
the claims adjustor’s experience. Ultimately, this full
sizing across the value chain made a step-change
impact on end-to-end handling time.

Buy selectively. Build strategically.
Matching the pace of innovation, many new startups and software offerings are entering the market,
leaving enterprises with a familiar question: “Buy
or build?” On the “buy” side, we see organizations
that are wary about investing in capabilities that
likely will eventually be available for a fraction of the
cost. These same organizations are also skeptical
of off-the-shelf solutions, unsure if the software will
perform at scale without significant customization.
As these solutions mature and prove their value, buy


-----

code components, which allows subsequent gen
AI efforts to build off the lessons learned from the
trailblazing ones. We have seen these components
accelerate delivery by 25 to 50 percent.

Throughout the past year, there has been an
endless stream of gen AI news and hype. The
coming year will likely be similar—but with a
growing focus on delivering real business value
to justify the billions in investment. From large
enterprises to pioneering start-ups, organizations


need to form their strategies around the
decades-old principles from digital and analytics
transformations. Organizations that get these
tried-and-tested learnings right will form lasting
strategic advantages against their competitors,
creating sticky customer experiences
and gaining market share in a challenging
macroeconomic environment.

If 2023 was a year of hype, then 2024 will be the
year of lasting impact at scale.


_This article was originally published November 29, 2023 on the QuantumBlack, AI by McKinsey_
_Medium blog._

Sam Bourton is a partner in the Lyon office; Ben Ellencweig is a senior partner in McKinsey’s Stamford office; Carlo Giovine
is a partner in the London office; and Stephen Xu is director of product management for Quantum Black, AI by McKinseyR&D in the Toronto office.

Copyright © 2023 McKinsey & Company. All rights reserved.


-----

**Appendix: Generative AI**
**solutions in action**

There’s plenty of hype around generative AI. But does the technology actually deliver?

Evidence already exists to show that the answer is yes, unmistakably. The gen AI solutions, or tools,
listed in the exhibit come from an extensive pool of work McKinsey has done over the past year. These
solutions can help you understand ways gen AI might fit within your organization. If you are intrigued
by the technology and want to explore gen AI applications further, please reach out and schedule a
session with our experts. During these sessions, we can provide in-depth demonstrations (either live
or via video) of the technology and discuss how it can be tailored to meet your specific organizational
needs (for more information, visit McKinsey.com/GenAI).

The gen AI landscape is evolving quickly. Don’t miss out on the opportunity to stay ahead of the game.

Exhibit
**Early solutions demonstrate the practical potential of generative AI.**

|Gen AI solution/tool|Description|
|---|---|
|Tech services|Document Q&A solution for ticket resolution in app- managed services projects to assist with user support queries|
|Media call agent|Tool that helps a global tech and media company generate fast, high-quality knowledge searches for agents serving customers during calls|
|Billing and revenue assurance|Knowledge synthesis agent that pulls data from diverse sources to assist with revenue cycle processes|
|Insight synthesis agent|Gen AI solution that helps a global information services company generate quicker and more meaningful insights into consumer behavior|
|Virtual agent|Tool to automate web research tasks, insight generation, and interaction with third-party APIs|
|Virtual analyst|Interactive chatbot that helps users easily extract insights from their personal data|
|Agent copilot|Tool providing real-time agent support and personalized customer recommendations|
|AI voice analytics|Application that understands call reasons and derives actionable insights to reduce demand, improve routing, and boost agent performance|
|Call center coaching|Tool that identifies opportunities for coaching call center employees on hard and soft skills|


-----

|Gen AI solution/tool|Description|
|---|---|
|Voice-to-voice automation|Voice bot that responds to customer questions with customized answers and/or quick automated tasks|
|Marketing adviser|Tool that summarizes social media posts for marketing insights|
|Sales.ai|End-to-end solution leveraging analytical and gen AI to identify leads and conduct customer outreach at scale|
|Ada|One-point solution for gen AI procurement use cases (eg, negotiation preparation, contract review, idea generation, category facts, market and supplier news)|
|Doc IQ|Tool that uses advanced deep learning techniques to analyze and extract data from contract documents and images and collect the data into a structured format|
|RFP generator|Tool that lets users input the context of a request for proposal (RFP) and follow up with chat prompts to generate new RFP content automatically|
|Procurement contract AI|Gen AI tool that reviews contracts, comparing terms and clauses against predefined best practices and McKinsey proprietary knowledge, providing instant insights|
|Coding copilot|AI assistant to generate and troubleshoot code|


-----

**Glossary**

Application programming interface (API) is a way to programmatically access (usually
external) models, datasets, or other pieces of software.

Artificial intelligence (AI) is the ability of software to perform tasks that traditionally require
human intelligence.

Artificial neural networks (ANNs) are composed of interconnected layers of software-based
calculators known as “neurons.” These networks can absorb vast amounts of input data and
process that data through multiple layers that extract and learn the data’s features.

Deep learning is a subset of machine learning that uses deep neural networks, which are
layers of connected “neurons” whose connections have parameters or weights that can be
trained. It is especially effective at learning from unstructured data such as images, text, and
audio.

Early and late scenarios are the extreme scenarios of our work-automation model. The
“earliest” scenario flexes all parameters to the extremes of plausible assumptions, resulting
in faster automation development and adoption; the “latest” scenario flexes all parameters in
the opposite direction. The reality is likely to fall somewhere between the two.

Fine-tuning is the process of adapting a pretrained foundation model to perform better in
a specific task. This entails a relatively short period of training on a labeled dataset, which
is much smaller than the dataset the model was initially trained on. This additional training
allows the model to learn and adapt to the nuances, terminology, and specific patterns found
in the smaller dataset.

Foundation models (FMs) are deep learning models trained on vast quantities of
unstructured, unlabeled data that can be used for a wide range of tasks out of the box or
adapted to specific tasks through fine-tuning. Examples of these models are GPT-4, PaLM,
DALL·E 2, and Stable Diffusion.

Generative AI is AI that is typically built using foundation models and has capabilities that
earlier AI did not have, such as the ability to generate content. Foundation models can also
be used for nongenerative purposes (for example, classifying user sentiment as negative or
positive based on call transcripts) while offering significant improvement over earlier models.
For simplicity, when we refer to generative AI, we include all foundation model use cases.

Graphics processing units (GPUs) are computer chips that were originally developed for
producing computer graphics (such as for video games) and are also useful for deep learning
applications. In contrast, traditional machine learning and other analyses usually run on
central processing units (CPUs), normally referred to as a computer’s “processor.”

Large language models (LLMs) make up a class of foundation models that can process
massive amounts of unstructured text and learn the relationships between words or portions
of words, known as tokens. This enables LLMs to generate natural-language text, performing


-----

tasks such as summarization or knowledge extraction. GPT-4 (which underlies ChatGPT) and
LaMDA (the model behind Bard) are examples of LLMs.

Machine learning (ML) is a subset of AI in which a model gains capabilities after it is trained
on, or shown, many example data points. Machine learning algorithms detect patterns and
learn how to make predictions and recommendations by processing data and experiences,
rather than by receiving explicit programming instruction. The algorithms also adapt and can
become more effective in response to new data and experiences.

Modality is a high-level data category such as numbers, text, images, video, and audio.

Productivity from labor is the ratio of GDP to total hours worked in the economy. Labor
productivity growth comes from increases in the amount of capital available to each worker,
the education and experience of the workforce, and improvements in technology.

Prompt engineering refers to the process of designing, refining, and optimizing input
prompts to guide a generative AI model toward producing desired (that is, accurate) outputs.

Self-attention, sometimes called intra-attention, is a mechanism that aims to mimic cognitive
attention, relating different positions of a single sequence to compute a representation of the
sequence.

Structured data is tabular data (for example, organized in tables, databases, or
spreadsheets) that can be used to train some machine learning models effectively.

Transformers are a relatively new neural network architecture that relies on self-attention
mechanisms to transform a sequence of inputs into a sequence of outputs while focusing its
attention on important parts of the context around the inputs. Transformers do not rely on
convolutions or recurrent neural networks.

Technical automation potential refers to the share of the work time that could be
automated. We assessed the technical potential for automation across the global economy
through an analysis of the component activities of each occupation. We used databases
published by institutions including the World Bank and the US Bureau of Labor Statistics to
break down about 850 occupations into approximately 2,100 activities, and we determined
the performance capabilities needed for each activity based on how humans currently
perform them.

Use cases are applications targeted to a specific business challenge that produce one or
more measurable outcomes. For example, in marketing, generative AI could be used to
generate creative content such as personalized emails.

Unstructured data lacks a consistent format or structure (for example, text, images, and
audio files) and typically requires more advanced techniques to extract insights.


-----

February 2024
Copyright © McKinsey & Company

www.McKinsey.com

@McKinsey

@McKinsey

@McKinsey


-----

