Generative AI Potential

June 2023
The economic
potential of
generative AI
The next productivity frontier
Authors
Michael Chui
Eric Hazan
Roger Roberts
Alex Singla
Kate Smaje
Alex Sukharevsky
Lareina Yee
Rodney Zemmel

ii The economic potential of generative AI: The next productivity frontier

Contents
Key insights
3
Chapter 1: Generative AI
as a technology catalyst
4
Glossary
6
Chapter 2: Generative AI use
cases across functions and
industries
8
Spotlight: Retail and
consumer packaged goods
27
Spotlight: Banking
28
Spotlight: Pharmaceuticals
and medical products
30
Chapter 3: The generative
AI future of work: Impacts
on work activities, economic
growth, and productivity
32
Chapter 4: Considerations
for businesses and society
48
Appendix
53
1
The economic potential of generative AI: The next productivity frontier

2 The economic potential of generative AI: The next productivity frontier

1. Generative AI’s impact on
productivity could add trillions
of dollars in value to the global
economy. Our latest research
estimates that generative AI could
add the equivalent of $2.6 trillion
to $4.4 trillion annually across the
63 use cases we analyzed—by
comparison, the United Kingdom’s
entire GDP in 2021 was $3.1 trillion.
This would increase the impact of
all artificial intelligence by 15 to
40 percent. This estimate would
roughly double if we include the
impact of embedding generative AI
into software that is currently used
for other tasks beyond those use
cases.
2. About 75 percent of the value that
generative AI use cases could
deliver falls across four areas:
Customer operations, marketing
and sales, software engineering,
and R&D. Across 16 business
functions, we examined 63 use
cases in which the technology
can address specific business
challenges in ways that produce
one or more measurable outcomes.
Examples include generative AI’s
ability to support interactions
with customers, generate creative
content for marketing and sales,
and draft computer code based on
natural-language prompts, among
many other tasks.
3. Generative AI will have a significant
impact across all industry sectors.
Banking, high tech, and life
sciences are among the industries
that could see the biggest impact
as a percentage of their revenues
from generative AI. Across the
banking industry, for example, the
technology could deliver value
equal to an additional $200 billion
to $340 billion annually if the use
cases were fully implemented. In
retail and consumer packaged
goods, the potential impact is also
significant at $400 billion to $660
billion a year.
4. Generative AI has the potential
to change the anatomy of work,
augmenting the capabilities of
individual workers by automating
some of their individual activities.
Current generative AI and other
technologies have the potential to
automate work activities that absorb
60 to 70 percent of employees’ time
today. In contrast, we previously
estimated that technology has the
potential to automate half of the
time employees spend working.1
The acceleration in the potential for
technical automation is largely due
to generative AI’s increased ability
to understand natural language,
which is required for work activities
that account for 25 percent of total
work time. Thus, generative AI has
more impact on knowledge work
associated with occupations that
have higher wages and educational
requirements than on other types
of work.
5. The pace of workforce
transformation is likely to
accelerate, given increases in the
potential for technical automation.
Our updated adoption scenarios,
including technology development,
economic feasibility, and diffusion
timelines, lead to estimates that
half of today’s work activities could
be automated between 2030 and
2060, with a midpoint in 2045, or
roughly a decade earlier than in our
previous estimates.
6. Generative AI can substantially
increase labor productivity across
the economy, but that will require
investments to support workers
as they shift work activities or
change jobs. Generative AI could
enable labor productivity growth
of 0.1 to 0.6 percent annually
through 2040, depending on the
rate of technology adoption and
redeployment of worker time
into other activities. Combining
generative AI with all other
technologies, work automation
could add 0.2 to 3.3 percentage
points annually to productivity
growth. However, workers will need
support in learning new skills, and
some will change occupations. If
worker transitions and other risks
can be managed, generative AI
could contribute substantively to
economic growth and support a
more sustainable, inclusive world.
7. The era of generative AI is just
beginning. Excitement over this
technology is palpable, and early
pilots are compelling. But a full
realization of the technology’s
benefits will take time, and leaders
in business and society still
have considerable challenges to
address. These include managing
the risks inherent in generative
AI, determining what new skills
and capabilities the workforce will
need, and rethinking core business
processes such as retraining and
developing new skills.
Key insights
3

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
Diffusion, and other generative AI tools that have captured current public attention are the
result of significant 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 classification 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.
1

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.2
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.3
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 2023.4
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.5
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.6
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 the potential for generative AI to create
value for companies and alter the workforce. The following sections share our initial findings.
5

Glossary
Application programming interface (API) is a way to programmatically access (usually
external) models, data sets, 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, and 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 data set, which
is much smaller than the data set 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 data set.
Foundation models (FM) 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 in this article, 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 are 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 worktime 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 targeted applications to a specific business challenge that produces one
or more measurable outcomes. For example, in marketing, generative AI could be used to
generate creative content such as personalized emails.
Unstructured data lack a consistent format or structure (for example, text, images, and audio
files) and typically require more advanced techniques to extract insights.
7

Generative AI is a step change in the evolution of artificial 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).
Generative AI use
cases across functions
and industries
2

The first lens scans use cases for generative AI that organizations could adopt. We define
a “use case” as a targeted application of generative AI to a specific 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 effectiveness of higher-quality content
at scale. We identified 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 benefits 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 artificial 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 first 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 affect labor productivity across all work currently done by
the global workforce.
Exhibit 1
The potential impact of generative AI can be evaluated through two lenses.
McKinsey & Company
Lens 1
Total economic
potential of 60-plus
organizational use
cases1
1
For 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.
Revenue
impacts of
use cases1
Cost impacts
of use cases
Lens 2
Labor productivity potential
across ~2,100 detailed work
activities performed by
global workforce
9

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
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).
Exhibit 2
Generative AI could create additional value potential above what
could be unlocked by other AI and analytics.
McKinsey & Company
AI’s potential impact on the global economy, $ trillion
1
Updated use case estimates from "Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
Advanced analytics,
traditional machine
learning, and deep
learning1
New generative
AI use cases
Total use
case-driven
potential
All worker productivity
enabled by generative
AI, including in use
cases
Total AI
economic
potential
11.0–17.7
13.6–22.1
17.1–25.6
2.6–4.4
6.1–7.9
~15–40%
incremental
economic impact
~35–70%
incremental
economic impact

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
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 (please
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.
Box 1
1
“Notes from the AI frontier: Applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
How we estimated the value potential of generative AI use cases
To assess the potential value of generative AI,
we updated a proprietary McKinsey database of
potential AI use cases and drew on the experience
of more than 100 experts in industries and their
business functions.1
Our updates examined
use cases of generative AI—specifically, how
generative AI techniques (primarily transformer-
based neural networks) can be used to solve
problems not well addressed by previous
technologies.
We analyzed only use cases for which generative
AI could deliver a significant improvement in the
outputs that drive key value. In particular, our
estimates of the primary value the technology
could unlock do not include use cases for which
the sole benefit would be its ability to use natural
language. For example, natural-language
capabilities would be the key driver of value in
a customer service use case but not in a use
case optimizing a logistics network, where value
primarily arises from quantitative analysis.
We then estimated the potential annual value
of these generative AI use cases if they were
adopted across the entire economy. For use
cases aimed at increasing revenue, such as some
of those in sales and marketing, we estimated
the economy-wide value generative AI could
deliver by increasing the productivity of sales and
marketing expenditures.
Our estimates are based on the structure of the
global economy in 2022 and do not consider the
value generative AI could create if it produced
entirely new product or service categories.
11

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 identified 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.
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.7
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.
Exhibit 3
Web <2023>
<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.
McKinsey & Company
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
Impact as a percentage of functional spend, %
Impact, $ billion
Marketing
Sales
Pricing
Customer operations
Corporate IT1
Product and R&D1
Software engineering
(for corporate IT)
(for product development)
Supply chain
Procurement management
Manufacturing
Legal
Risk and compliance
Strategy
Finance
Talent and organization (incl HR)
0 10 20 30 40
0
100
200
300
400
500
Represent ~75% of total annual impact of generative 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 work week, 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 specific use
cases, the technology
could drive value across
an entire organization
by revolutionizing
internal knowledge
management systems.
13

How customer operations
could be transformed
Customer self-service interactions
Customer interacts with a humanlike chatbot that
delivers immediate, personalized responses to
complex inquiries, ensuring a consistent brand
voice regardless of customer language or location.
Customer–agent interactions
Human agent uses AI-developed call scripts and
receives real-time assistance and suggestions for
responses during phone conversations, instantly
accessing relevant customer data for tailored and
real-time information delivery.
Agent self-improvement
Agent receives a summarization of the conversation in
a few succinct points to create a record of customer
complaints and actions taken.
Agent uses automated, personalized insights generated
by AI, including tailored follow-up messages or
personalized coaching suggestions.

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.8
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.
15

How marketing and sales
Strategization
Sales and marketing professionals efficiently
gather market trends and customer information
from unstructured data sources (for example,
social media, news, research, product information,
and customer feedback) and draft effective
marketing and sales communications.
Awareness
Customers see campaigns tailored
to their segment, language, and
demographic.
Consideration
Customers can access comprehensive information,
comparisons, and dynamic recommendations, such as
personal “try ons.”

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, 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.
Conversion
Virtual sales representatives enabled by generative
AI emulate humanlike qualities—such as empathy,
personalized communication, and natural language
processing—to build trust and rapport with
customers.
Retention
Customers are more likely to be retained with
customized messages and rewards, and they can
interact with AI-powered customer-support chatbots
that manage the relationship proactively, with fewer
escalations to human agents.
17

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 ecommerce 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.
19

How software engineering
Inception and planning
Software engineers and product managers use
generative AI to assist in analyzing, cleaning, and labeling
large volumes of data, such as user feedback, market
trends, and existing system logs.
System design
Engineers use generative AI to create multiple IT
architecture designs and iterate on the potential
configurations, accelerating system design, and
allowing faster time to market.
Coding
Engineers are assisted by AI tools that can code,
reducing development time by assisting with drafts,
rapidly finding prompts, and serving as an easily
navigable knowledge base.
Testing
Engineers employ algorithms that can enhance
functional and performance testing to ensure
quality and can generate test cases and test data
automatically.

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.9
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.10
Maintenance
Engineers use AI insights on system logs, user feedback,
and performance data to help diagnose issues,
suggest fixes, and predict other high-priority areas of
improvement.
21

How product R&D
Early research analysis
Researchers use generative AI to enhance market reporting,
ideation, and product or solution drafting.
Virtual design
Researchers use generative AI to generate
prompt-based drafts and designs, allowing them
to iterate quickly with more design options.
Virtual simulations
Researchers accelerate and optimize the
virtual simulation phase if combined with new
deep learning generative design techniques.
Physical test planning
Researchers optimize test cases for more
efficient testing, reducing the time required
for physical build and testing.
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.
Box 2
Deep learning surrogates
Product design in industries producing
physical products often involves physics-
based 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 time-
and 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 compute-
intensive 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)physics-
based 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.
23

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.
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.11
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).
Across 63 use cases,
generative AI has the
potential to generate
$2.6 trillion to $4.4 trillion
in value across industries.

Exhibit 4
Generative AI use cases will have different impacts on business functions
across industries.
McKinsey & Company
Administrative and
professional services
Advanced electronics
and semiconductors
Advanced manufacturing3
Agriculture
Banking
Basic materials
Chemical
Construction
Consumer packaged goods
Education
Energy
Healthcare
High tech
Insurance
Media and entertainment
Pharmaceuticals and
medical products
Public and social sector
Real estate
Retail4
Telecommunications
Travel, transport, and logistics
Total,
$ billion
150–250
100–170
170–290
40–70
200–340
120–200
80–140
90–150
160–270
120–230
150–240
150–260
240–460
50–70
60–110
60–110
70–110
110–180
240–390
60–100
180–300
Total, % of
industry
revenue
0.9–1.4
1.3–2.3
1.4–2.4
0.6–1.0
2.8–4.7
0.7– 1.2
0.8–1.3
0.7–1.2
1.4–2.3
2.2–4.0
1.0– 1.6
1.8–3.2
4.8–9.3
1.8– 2.8
1.5– 2.6
2.6–4.5
0.5–0.9
1.0–1.7
1.2–1.9
2.3–3.7
1.2–2.0
Generative AI productivity
impact by business functions¹
M
a
r
k
e
t
i
n
g
a
n
d
s
a
l
e
s
C
u
s
t
o
m
e
r
o
p
e
r
a
t
i
o
n
s
P
r
o
d
u
c
t
a
n
d
R
&
D
S
o
f
t
w
a
r
e
e
n
g
i
n
e
e
r
i
n
g
S
u
p
p
l
y
c
h
a
i
n
a
n
d
o
p
e
r
a
t
i
o
n
s
R
i
s
k
a
n
d
l
e
g
a
l
S
t
r
a
t
e
g
y
a
n
d
fi
n
a
n
c
e
C
o
r
p
o
r
a
t
e
I
T
2
T
a
l
e
n
t
a
n
d
o
r
g
a
n
i
z
a
t
i
o
n
2,600–4,400
Note: Figures may not sum to 100%, because of rounding.
1
Excludes implementation costs (eg, training, licenses).
2
Excluding software engineering.
3
Includes aerospace, defense, and auto manufacturing.
4
Including 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
760–
1,200
340–
470
230–
420
580–
1,200
280–
530
180–
260
120–
260
40–
50
60–
90
Low impact High impact
25

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 (Exhibit 5).
In the banking industry, generative AI has the potential to improve on efficiencies already
delivered by artificial intelligence by taking on lower-value tasks in risk management, such
as required reporting, monitoring regulatory developments, and collecting data. In the life
sciences industry, generative AI is poised to make significant contributions to drug discovery
and development.
We share our detailed analysis of these industries in the following industry spotlights.
Exhibit 5
Generative AI could deliver significant value when deployed in some use cases
across a selection of top industries.
McKinsey & Company
Selected examples of key use cases for main functional
value drivers (nonexhaustive)
Value
potential,
as % of
operating
profits1
Product R&D,
software
engineering
Customer
operations
Marketing
and sales
Other
functions
Retail
and
consumer
packaged
goods2
400–660
(1–2%)
27–44 Consumer research
Accelerate consumer
research by testing
scenarios, and
enhance customer
targeting by creating
“synthetic customers”
to practice with
Augmented
reality–assisted
customer support
Rapidly inform the
workforce in real
time about the status
of products and
consumer
preferences
Assist copy writing
for marketing
content creation
Accelerate writing of
copy for marketing
content and
advertising scripts
Procurement
suppliers
process
enhancement
Draft playbooks
for negotiating
with suppliers
Pharma
and
medical
products
60–110
(3–5%)
15–25 Research and
drug discovery
Accelerate the
selection of proteins
and molecules best
suited as candidates
for new drug
formulation
Customer
documentation
generation
Draft medication
instructions and risk
notices for drug
resale
Generate content
for commercial
representatives
Prepare scripts for
interactions with
physicians
Contract
generation
Draft legal
documents
incorporating
specific
regulatory
requirements
Banking 200–340
(3–5%)
9–15 Legacy code
conversion
Optimize migration
of legacy
frameworks with
natural-language
translation
capabilities
Customer
emergency
interactive voice
response (IVR)
Partially automate,
accelerate, and
enhance resolution
rate of customer
emergencies through
generative
AI–enhanced IVR
interactions (eg, for
credit card losses)
Custom retail
banking offers
Push personalized
marketing and sales
content tailored for
each client of the
bank based on
profile and history
(eg, personalized
nudges), and
generate alternatives
for A/B testing
Risk model
documentation
Create model
documentation,
and scan for
missing
documentation
and relevant
regulatory
updates
Total value
potential
per industry,
$ billion (%
of industry
revenue)
Value potential
of function for
the industry Low
High
¹Operating profit based on average profitability of selected industries in the 2020–22 period.
2Includes auto retail.

Spotlight: Retail and CPG
Generative AI could change the game for retail
and consumer packaged goods companies
1
Vehicular retail is included as part of our overall retail analysis.
The technology could generate value for
the retail and consumer packaged goods
(CPG) industry by increasing productivity
by 1.2 to 2.0 percent of annual revenues,
or an additional $400 billion to $660 bil-
lion.1
To streamline processes, generative
AI could automate key functions such as
customer service, marketing and sales,
and inventory and supply chain manage-
ment.
Technology has played an essen-
tial role in the retail and CPG indus-
tries for decades. Traditional AI and
advanced-analytics solutions have
helped companies manage vast pools
of data across large numbers of SKUs,
expansive supply chain and warehousing
networks, and complex product catego-
ries such as consumables.
In addition, the industries are heavily
customer facing, which offers opportu-
nities for generative AI to complement
previously existing artificial intelli-
gence. For example, generative AI’s
ability to personalize offerings could
optimize marketing and sales activities
already handled by existing AI solutions.
Similarly, generative AI tools excel at data
management and could support existing
AI-driven pricing tools. Applying gener-
ative AI to such activities could be a step
toward integrating applications across a
full enterprise.
Generative AI is already at work in some
retail and CPG companies:
Reinvention of the customer
interaction pattern
Consumers increasingly seek customiza-
tion in everything from clothing and cos-
metics to curated shopping experiences,
personalized outreach, and food—and
generative AI can improve that expe-
rience. Generative AI can aggregate
market data to test concepts, ideas, and
models. Stitch Fix, which uses algorithms
to suggest style choices to its custom-
ers, has experimented with DALL·E to
visualize products based on customer
preferences regarding color, fabric, and
style. Using text-to-image generation,
the company’s stylists can visualize an
article of clothing based on a consumer’s
preferences and then identify a similar
article among Stitch Fix’s inventory.
Retailers can create applications that
give shoppers a next-generation experi-
ence, creating a significant competitive
advantage in an era when customers
expect to have a single natural-language
interface help them select products. For
example, generative AI can improve the
process of choosing and ordering ingre-
dients for a meal or preparing food—
imagine a chatbot that could pull up the
most popular tips from the comments
attached to a recipe. There is also a big
opportunity to enhance customer value
management by delivering personalized
marketing campaigns through a chatbot.
Such applications can have human-like
conversations about products in ways
that can increase customer satisfaction,
traffic, and brand loyalty. Generative
AI offers retailers and CPG companies
many opportunities to cross-sell and
upsell, collect insights to improve prod-
uct offerings, and increase their cus-
tomer base, revenue opportunities, and
overall marketing ROI.
Accelerating the creation
of value in key areas
Generative AI tools can facilitate copy
writing for marketing and sales, help
brainstorm creative marketing ideas,
expedite consumer research, and accel-
erate content analysis and creation. The
potential improvement in writing and
visuals can increase awareness and
improve sales conversion rates.
Rapid resolution and enhanced
insights in customer care
The growth of e-commerce also elevates
the importance of effective consumer
interactions. Retailers can combine
existing AI tools with generative AI to
enhance the capabilities of chatbots,
enabling them to better mimic the
interaction style of human agents—for
example, by responding directly to a
customer’s query, tracking or cancel-
ing an order, offering discounts, and
upselling. Automating repetitive tasks
allows human agents to devote more
time to handling complicated customer
problems and obtaining contextual infor-
mation.
Disruptive and creative innovation
Generative AI tools can enhance the
process of developing new versions
of products by digitally creating new
designs rapidly. A designer can generate
packaging designs from scratch or gen-
erate variations on an existing design.
This technology is developing rapidly and
has the potential to add text-to-video
generation.
Additional factors to consider
As retail and CPG executives explore
how to integrate generative AI in their
operations, they should keep in mind
several factors that could affect their
ability to capture value from the technol-
ogy.
External inference. Generative AI has
increased the need to understand
whether generated content is based on
fact or inference, requiring a new level of
quality control.
Adversarial attacks. Foundation models
are a prime target for attack by hackers
and other bad actors, increasing the vari-
ety of potential security vulnerabilities
and privacy risks.
To address these concerns, retail and
CPG companies will need to strate-
gically keep humans in the loop and
ensure security and privacy are top
considerations for any implementation.
Companies will need to institute new
quality checks for processes previous-
ly handled by humans, such as emails
written by customer reps, and per-
form more-detailed quality checks on
AI-assisted processes such as product
design.
27

Spotlight: Banking
1
“Building the AI bank of the future,” McKinsey, May 2021.
2
McKinsey’s Global Banking Annual Review, December 1, 2022.
3
Akhil Babbar, Raghavan Janardhanan, Remy Paternoster, and Henning Soller, “Why most digital banking transformations fail—and how to flip the odds,” McKinsey,
April 11, 2023.
4
Hugh Son, “Morgan Stanley is testing an OpenAI-powered chatbot for its 16,000 financial advisors,” CNBC, March 14, 2023.
Banks could realize substantial
value from generative AI
Generative AI could have a significant
impact on the banking industry, gener-
ating value from increased productivity
of 2.8 to 4.7 percent of the industry’s
annual revenues, or an additional $200
billion to $340 billion. On top of that
impact, the use of generative AI tools
could also enhance customer satis-
faction, improve decision making and
employee experience, and decrease
risks through better monitoring of fraud
and risk.
Banking, a knowledge and technolo-
gy-enabled industry, has already bene-
fited significantly from previously exist-
ing applications of artificial intelligence
in areas such as marketing and custom-
er operations.1
Generative AI applica-
tions could deliver additional benefits,
especially because text modalities are
prevalent in areas such as regulations
and programming language, and the
industry is customer facing, with many
B2C and small-business customers.2
Several characteristics position the
industry for the integration of genera-
tive AI applications:
— Sustained digitization efforts along
with legacy IT systems. Banks
have been investing in technology
for decades, accumulating a
significant amount of technical debt
along with a siloed and complex IT
architecture.3
— Large customer-facing workforces.
Banking relies on a large number
of service representatives such
as call-center agents and wealth
management financial advisers.
— A stringent regulatory environment.
As a heavily regulated industry,
banking has a substantial number of
risk, compliance, and legal needs.
— White-collar industry. Generative
AI’s impact could span the
organization, assisting all
employees in writing emails,
creating business presentations,
and other tasks.
On the move
Banks have started to grasp the poten-
tial of generative AI in their front lines
and in their software activities. Early
adopters are harnessing solutions such
as ChatGPT as well as industry-specific
solutions, primarily for software and
knowledge applications. Three uses
demonstrate its value potential to the
industry:
A virtual expert to augment
employee performance
A generative AI bot trained on pro-
prietary knowledge such as policies,
research, and customer interaction
could provide always-on, deep techni-
cal support. Today, frontline spending
is dedicated mostly to validating offers
and interacting with clients, but giv-
ing frontline workers access to data
as well could improve the customer
experience. The technology could also
monitor industries and clients and
send alerts on semantic queries from
public sources. For example, Morgan
Stanley is building an AI assistant using
GPT-4, with the aim of helping tens of
thousands of wealth managers quickly
find and synthesize answers from a
massive internal knowledge base.4
The
model combines search and content
creation so wealth managers can find
and tailor information for any client at
any moment.
One European bank has leveraged gen-
erative AI to develop an environmental,
social, and governance (ESG) virtual
expert by synthesizing and extracting
from long documents with unstruc-
tured information. The model answers
complex questions based on a prompt,
identifying the source of each answer
and extracting information from pic-
tures and tables.
Generative AI could reduce the signifi-
cant costs associated with back-office
operations. Such customer-facing
chatbots could assess user requests
and select the best service expert to
address them based on characteristics
such as topic, level of difficulty, and
type of customer. Through generative
AI assistants, service professionals
could rapidly access all relevant infor-
mation such as product guides and
policies to instantaneously address
customer requests.
Code acceleration to reduce tech
debt and deliver software faster
Generative AI tools are useful for soft-
ware development in four broad cate-
gories. First, they can draft code based
on context via input code or natural
language, helping developers code
more quickly and with reduced friction
while enabling automatic translations
and no- and low-code tools. Second,
such tools can automatically generate,
prioritize, run, and review different
code tests, accelerating testing and
increasing coverage and effectiveness.
Third, generative AI’s natural-language
translation capabilities can optimize
the integration and migration of legacy
frameworks. Last, the tools can review
code to identify defects and inefficien-
cies in computing. The result is more
robust, effective code.

Production of tailored
content at scale
Generative AI tools can draw on existing
documents and data sets to substan-
tially streamline content generation.
These tools can create personalized
marketing and sales content tailored
to specific client profiles and histories
as well as a multitude of alternatives
for A/B testing. In addition, generative
AI could automatically produce model
documentation, identify missing docu-
mentation, and scan relevant regulatory
updates to create alerts for relevant
shifts.
Factors for banks to consider
When exploring how to integrate gen-
erative AI into operations, banks can be
mindful of a number of factors:
— The level of regulation for different
processes. These vary from
unregulated processes such
as customer service to heavily
regulated processes such as credit
risk scoring.
— Type of end user. End users vary
widely in their expectations and
familiarity with generative AI—for
example, employees compared with
high-net-worth clients.
— Intended level of work automation.
AI agents integrated through APIs
could act nearly autonomously
or as copilots, giving real-time
suggestions to agents during
customer interactions.
— Data constraints. While public data
such as annual reports could be
made widely available, there would
need to be limits on identifiable
details for customers and other
internal data.
A generative AI bot trained
on proprietary knowledge
such as policies, research,
and customer interaction
could provide always-on,
deep technical support.
29

Spotlight: Pharmaceuticals and medical products
Generative AI deployment could unlock
potential value equal to 2.6 to 4.5 percent of
annual revenues across the pharmaceutical
and medical-product industries
1
Research and development in the pharmaceutical industry, Congressional Budget Office, April 2021.
Our analysis finds that generative AI
could have a significant impact on the
pharmaceutical and medical-product
industries—from $60 billion to $110 bil-
lion annually. This big potential reflects
the resource-intensive process of dis-
covering new drug compounds. Pharma
companies typically spend approximate-
ly 20 percent of revenues on R&D,1
and
the development of a new drug takes an
average of ten to 15 years.
With this level of spending and time-
line, improving the speed and quality
of R&D can generate substantial value.
For example, lead identification—a
step in the drug discovery process in
which researchers identify a molecule
that would best address the target for
a potential new drug—can take several
months even with “traditional” deep
learning techniques. Foundation models
and generative AI can enable organiza-
tions to complete this step in a matter of
weeks.
Generative AI use cases
aligned to industry needs
Drug discovery involves narrowing the
universe of possible compounds to those
that could effectively treat specific con-
ditions. Generative AI’s ability to process
massive amounts of data and model
options can accelerate output across
several use cases:
Improve automation of
preliminary screening
In the lead identification stage of drug
development, scientists can use founda-
tion models to automate the preliminary
screening of chemicals in the search for
those that will produce specific effects
on drug targets. To start, thousands of
cell cultures are tested and paired with
images of the corresponding experi-
ment. Using an off-the-shelf foundation
model, researchers can cluster similar
images more precisely than they can
with traditional models, enabling them to
select the most promising chemicals for
further analysis during lead optimization.
Enhance indication finding
An important phase of drug discovery
involves the identification and prioriti-
zation of new indications—that is, dis-
eases, symptoms, or circumstances that
justify the use of a specific medication
or other treatment, such as a test, pro-
cedure, or surgery. Possible indications
for a given drug are based on a patient
group’s clinical history and medical
records, and they are then prioritized
based on their similarities to established
and evidence-backed indications.
Researchers start by mapping the
patient cohort’s clinical events and
medical histories—including potential
diagnoses, prescribed medications,
and performed procedures—from real-
world data. Using foundation models,
researchers can quantify clinical events,
establish relationships, and measure the
similarity between the patient cohort
and evidence-backed indications. The
result is a short list of indications that
have a better probability of success in
clinical trials because they can be more
accurately matched to appropriate
patient groups.
Pharma companies that have used this
approach have reported high success
rates in clinical trials for the top five indi-
cations recommended by a foundation
model for a tested drug. This success
has allowed these drugs to progress
smoothly into Phase 3 trials, significantly
accelerating the drug development pro-
cess.
Additional factors to consider
Before integrating generative AI into
operations, pharma executives should
be aware of some factors that could limit
their ability to capture its benefits:
— The need for a human in the loop.
Companies may need to implement
new quality checks on processes
that shift from humans to generative
AI, such as representative-generated
emails, or more detailed quality
checks on AI-assisted processes,
such as drug discovery. The
increasing need to verify whether
generated content is based on fact
or inference elevates the need for a
new level of quality control.
— Explainability. A lack of transparency
into the origins of generated content
and traceability of root data could
make it difficult to update models
and scan them for potential risks;
for instance, a generative AI solution
for synthesizing scientific literature
may not be able to point to the
specific articles or quotes that led
it to infer that a new treatment is
very popular among physicians. The
technology can also “hallucinate,”
or generate responses that are
obviously incorrect or inappropriate
for the context. Systems need to be
designed to point to specific articles
or data sources, and then do human-
in-the-loop checking.
— Privacy considerations. Generative
AI’s use of clinical images and
medical records could increase the
risk that protected health information
will leak, potentially violating
regulations that require pharma
companies to protect patient privacy.

In this chapter, we have estimated the organizational value generative AI could deliver
through use cases across industries and business functions, but the technology’s potential
is much greater. As it is embedded into tools used by every knowledge worker, its additional
impact may be more diffuse but no less valuable than that associated with these use cases.
Companies need to find ways to maximize the value created by the generative AI they deploy
while also taking care to monitor and manage its impact on their workforces and society at
large.
31

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
The generative AI future
of work: Impacts on
work activities, economic
growth, and productivity
3

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.12
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 (see Box 3, “About the research”).
Labor productivity will
have to accelerate to
achieve economic growth
and enhance prosperity.
33

Box 3
About the research
This analysis builds on the methodology
we established in 2017. We began by
examining the US Bureau of Labor
Statistics O*Net breakdown of about
850 occupations into roughly 2,100
detailed work activities. For each of
these activities, we scored the level of
capability necessary to successfully
perform the activity against a set of 18
capabilities that have the potential for
automation (exhibit).
We also surveyed experts in the
automation of each of these capabilities
to estimate automation technologies’
current performance level against each
of these capabilities, as well as how
the technology’s performance might
advance over time. Specifically, this
year, we updated our assessments of
technology’s performance in cognitive,
language, and social and emotional
capabilities based on a survey of
generative AI experts.
Based on these assessments of the
technical automation potential of each
detailed work activity at each point in
time, we modeled potential scenarios for
the adoption of work automation around
the world. First, we estimated a range
of time to implement a solution that
could automate each specific detailed
work activity, once all the capability
requirements were met by the state of
technology development. Second, we
estimated a range of potential costs
for this technology when it is first
introduced, and then declining over
time, based on historical precedents.
We modeled the beginning of adoption
for a specific detailed work activity in a
particular occupation in a country (for 47
countries, accounting for more than 80
percent of the global workforce) when
the cost of the automation technology
reaches parity with the cost of human
labor in that occupation.
Based on a historical analysis of
various technologies, we modeled a
range of adoption timelines from eight
to 27 years between the beginning
of adoption and its plateau, using
sigmoidal curves (S-curves). This range
implicitly accounts for the many factors
that could affect the pace at which
adoption occurs, including regulation,
levels of investment, and management
decision making within firms.
The modeled scenarios create a
time range for the potential pace of
automating current work activities.
The “earliest” scenario flexes all
parameters to the extremes of plausible
assumptions, resulting in faster
automation development and adoption,
and the “latest” scenario flexes all
parameters in the opposite direction.
The reality is likely to fall somewhere
between the two.
Exhibit
Our analysis assesses the potential for technical automation
across some 2,100 activities and 18 capabilities.
McKinsey & Company
Source: McKinsey Global Institute analysis
~850 occupations ~2,100 activities assessed
across all occupations
Capability requirements
Sensory
• Sensory perception
Cognitive
• Retrieving information
• Recognizing known
patterns and categories
(supervised learning)
• Generating novel patterns
and categories
• Logical reasoning and
problem solving
• Optimizing and planning
• Creativity
• Articulating/display output
• Coordination with multiple
agents
Example: Retail activities
• Answer questions about
products and services
• Greet customers
• Clean and maintain work areas
• Demonstrate product features
• Process sales and transactions
Examples
Physical
• Fine motor skills and
dexterity
• Gross motor skills
• Navigation
• Mobility
Natural-language
processing
• Understanding natural
language
• Generating natural language
Social
• Social and emotional
sensing
• Social and emotional
reasoning
• Social and emotional output
Retail
salespeople
Food and beverage
service workers
Health
practitioners
Teachers

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 6). For example, MGI
previously identified 2027 as the earliest year when median human performance for natural-
language understanding might be achieved in technology, but in this new analysis, the
corresponding point is 2023.
Exhibit 6
As a result of generative AI, experts assess that technology could achieve human-
level performance in some technical capabilities sooner than previously thought.
McKinsey & Company
Technical capabilities, level of human performance achievable by technology
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
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
Estimates post-recent
generative AI developments (2023)¹
Estimates pre-generative AI (2017)¹ Median Top quartile
Median Top quartile Line represents range
of expert estimates
35

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 60–70 percent. The technical
potential curve is quite steep because of the acceleration in generative AI’s natural-language
capabilities (Exhibit 7).
Interestingly, the range of times between the early and late scenarios has compressed
compared with the expert assessments in 2017, reflecting a greater confidence that higher
levels of technological capabilities will arrive by certain time periods.
Adoption lags behind technical automation potential
Our analysis of adoption scenarios accounts for the time required to integrate technological
capabilities into solutions that can automate individual work activities; the cost of these
technologies compared with that of human labor in different occupations and countries
around the world; and the time it has taken for technologies to diffuse across the economy.
With the acceleration in technical automation potential that generative AI enables, our
scenarios for automation adoption have correspondingly accelerated. These scenarios
encompass a wide range of outcomes, given that the pace at which solutions will be
developed and adopted will vary based on decisions that will be made on investments,
Exhibit 7
The advent of generative AI has pulled forward the potential for
technical automation.
McKinsey & Company
Technical automation potentials by scenario, %
Time spent on
current work
activities1
1
Includes 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.
2
Early and late scenarios reflect the ranges provided by experts (see Exhibit 6).
2020 2030 2040 2050 2060
50
60
70
80
90
100
Updated early scenario
including generative AI2
Updated late scenario
including generative AI2
2017 early scenario2
2017 late scenario2
2023

deployment, and regulation, among other factors. But they give an indication of the degree to
which the activities that workers do each day may shift.
As an example of how this might play out in a specific occupation, consider postsecondary
English language and literature teachers, whose detailed work activities include preparing
tests and evaluating student work. With generative AI’s enhanced natural-language
capabilities, more of these activities could be done by machines, perhaps initially to create
a first draft that is edited by teachers but perhaps eventually with far less human editing
required. This could free up time for these teachers to spend more time on other work
activities, such as guiding class discussions or tutoring students who need extra assistance.
Our previously modeled adoption scenarios suggested that 50 percent of time spent on 2016
work activities would be automated sometime between 2035 and 2070, with a midpoint
scenario around 2053. Our updated adoption scenarios, which account for developments in
generative AI, models the time spent on 2023 work activities reaching 50 percent automation
between 2030 and 2060, with a midpoint of 2045—an acceleration of roughly a decade
compared with the previous estimate (Exhibit 8).13
Exhibit 8
The midpoint scenario at which automation adoption could reach 50 percent
of time spent on current work activities has accelerated by a decade.
McKinsey & Company
Global automation of time spent on current work activities,1
%
1
Includes 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.
2
Early scenario: aggressive scenario for all key model parameters (technical automation potential, integration timelines, economic feasibility, and technology
diffusion rates.).
3
Late scenario: parameters are set for later adoption potential.
2020 2030 2040 2050 2060 2070 2080 2090
Updated early scenario including
generative AI2
Updated late scenario including
generative AI3
2017 early scenario2
2017 late scenario3
100
0
50%
20
40
60
80
Midpoint
2017
Midpoint
updated
37

Different countries, different pace of adoption
Adoption is also likely to be faster in developed countries, where wages are higher and
thus the economic feasibility of adopting automation occurs earlier. Even if the potential for
technology to automate a particular work activity is high, the costs required to do so have to
be compared with the cost of human wages. In countries such as China, India, and Mexico,
where wage rates are lower, automation adoption is modeled to arrive more slowly than in
higher-wage countries (Exhibit 9).
Our analyses of generative AI’s impact on work activities and the pace of automation adoption
rely on several assumptions and sensitivities (see Box 4, “Limitations of our analyses, key
assumptions, and sensitivities”).
Exhibit 9
Automation adoption is likely to be faster in developed economies, where
higher wages will make it economically feasible sooner.
McKinsey & Company
Automation adoption by scenario for select countries, %
1
Early scenario: aggressive scenario for all key model parameters (technical automation potential, integration timelines, economic feasibility, and technology
diffusion rates.).
2
Late scenario: parameters are set for the later adoption potential.
China
Germany France India
Japan Mexico
United States
100
0
20
40
60
80
100
0
20
40
60
80
Early scenario¹ Late scenario²
50%
50%

Box 4
1
David Autor et al., New frontiers: The origins and content of new work, 1940–2018, National Bureau of Economic Research working paper number 30389, August
2022; Jeffrey Lin, “Technological adaptation, cities, and new work,” Review of Economics and Statistics, May 2011, Volume 93, Number 2.
Limitations of our analyses, key assumptions, and sensitivities
This analysis considers the potential for
automation only of current work activities
and occupations. It does not account
for how those work activities may shift
over time or forecast new activities and
occupations.1
Also, the analysis accounts
solely for first-order effects. It does not
take into account how labor rates could
change, and it does not model changes
in labor force participation rates or other
general equilibrium effects. That said,
while these models account for the time
it may take for technology to be adopted
across an economy, technologies could
be adopted much more rapidly in an
individual organization. Other research
may reach different conclusions.
Our assessments of technology
capabilities are based on the best
estimates of experts involved in
developing automation technologies.
These assessments could change over
time, as they have changed since 2017.
The technology adoption curves
are based on historical findings that
technologies take eight to 27 years from
commercial availability to reach a plateau
in adoption. Some argue that the adoption
of generative AI will be faster due to the
ease of deploying these technologies.
That said, the case for a minimum of
eight years in our earliest scenario for
reaching a global plateau in adoption
accounts for the pace of adoption of other
technologies that have arguably had a
faster adoption potential—for example,
social networking as a consumer
technology that faced no barriers in
enterprise change management. Our
scenario also accounts for the significant
role of small and midsize enterprises
around the world, in addition to the
challenges of incorporating and managing
change in larger organizations.
In addition, this analysis does not assume
that the scale of work automation
equates directly to job losses. Like other
technologies, generative AI typically
enables individual activities within
occupations to be automated, not entire
occupations. Historically, the activities
in many occupations have shifted over
time as certain activities are automated.
However, organizations may decide
to realize the benefits of increased
productivity by reducing employment
in some job categories, a possibility we
cannot rule out.
Generative AI is likely to
have the biggest impact on
knowledge work, particularly
activities involving decision
making and collaboration,
which previously had the lowest
potential for automation.
39

Generative AI’s potential impact on knowledge work
Previous generations of automation technology were particularly effective at automating
data management tasks related to collecting and processing data. Generative AI’s natural-
language capabilities increase the automation potential of these types of activities somewhat.
But its impact on more physical work activities shifted much less, which isn’t surprising
because its capabilities are fundamentally engineered to do cognitive tasks.
As a result, generative AI is likely to have the biggest impact on knowledge work, particularly
activities involving decision making and collaboration, which previously had the lowest
potential for automation (Exhibit 10). Our estimate of the technical potential to automate
the application of expertise jumped 34 percentage points, while the potential to automate
management and develop talent increased from 16 percent in 2017 to 49 percent in 2023.
Generative AI’s ability to understand and use natural language for a variety of activities and
tasks largely explains why automation potential has risen so steeply. Some 40 percent of
the activities that workers perform in the economy require at least a median level of human
understanding of natural language.
Exhibit 10
Generative AI could have the biggest impact on collaboration and the application
of expertise, activities that previously had a lower potential for automation.
McKinsey & Company
Overall technical automation potential, comparison in midpoint scenarios, % in 2023
Note: Figures may not sum, because of rounding.
1
Previous assessment of work automation before the rise of generative AI.
2
Applying expertise to decision making, planning, and creative tasks.
3
Managing and developing people.
4
Performing physical activities and operating machinery in unpredictable environments.
5
Performing physical activities and operating machinery in predictable environments.
58.5
49.0
45.0
90.5
79.0
46.0
73.0
24.5
15.5
24.0
73.0
68.0
45.5
72.5
Activity groups
Decision
making and
collaboration
Data
management
Physical
Applying expertise²
Managing³
Interfacing with
stakeholders
Processing data
Collecting data
Performing unpredictable
physical work⁴
Performing predictable
physical work⁵
Without generative AI1
With generative AI

As a result, many of the work activities that involve communication, supervision,
documentation, and interacting with people in general have the potential to be automated
by generative AI, accelerating the transformation of work in occupations such as education
and technology, for which automation potential was previously expected to emerge
later (Exhibit 11).
Exhibit 11
Advances in technical capabilities could have the most impact on activities
performed by educators, professionals, and creatives.
McKinsey & Company
Impact of generative AI on technical automation potential in midpoint scenario, 2023
Note: Figures may not sum, because of rounding.
¹Previous assessment of work automation before the rise of generative AI.
2Includes data from 47 countries, representing about 80% of employment across the world.
Educator and workforce training
Business and legal
professionals
STEM professionals
Community services
Creatives and arts management
Office support
Managers
Health professionals
Customer service and sales
Property maintenance
Health aides, technicians,
and wellness
Production work
Food services
Transportation services
Mechanical installation
and repair
Agriculture
Builders
Total
54
62
57
65
53
87
44
43
57
38
43
82
78
49
67
63
53
63
15
32
28
39
28
66
27
29
45
29
34
73
70
42
61
59
49
51
Occupation group
With generative AI
Without generative AI¹ Overall technical automation potential,
comparison in midpoint scenarios,
% in 2023
Share of global
employment,2 %
4
5
3
3
1
9
3
2
10
4
3
12
5
3
4
21
7
100
41

Labor economists have often noted that the deployment of automation technologies tends
to have the most impact on workers with the lowest skill levels, as measured by educational
attainment, or what is called skill biased. We find that generative AI has the opposite pattern—
it is likely to have the most incremental impact through automating some of the activities of
more-educated workers (Exhibit 12).
Another way to interpret this result is that generative AI will challenge the attainment of
multiyear degree credentials as an indicator of skills, and others have advocated for taking
a more skills-based approach to workforce development in order to create more equitable,
efficient workforce training and matching systems.14
Generative AI could still be described as
skill-biased technological change, but with a different, perhaps more granular, description of
skills that are more likely to be replaced than complemented by the activities that machines
can do.
Previous generations of automation technology often had the most impact on occupations
with wages falling in the middle of the income distribution. For lower-wage occupations,
making a case for work automation is more difficult because the potential benefits of
automation compete against a lower cost of human labor. Additionally, some of the tasks
performed in lower-wage occupations are technically difficult to automate—for example,
manipulating fabric or picking delicate fruits. Some labor economists have observed a
Exhibit 12
Generative AI increases the potential for technical automation most in
occupations requiring higher levels of educational attainment.
McKinsey & Company
Impact of generative AI on technical automation potential in midpoint scenario, 2023
57
60
62
64
64
63
28
36
45
48
51
54
Education level
Master’s, PhD, or higher
With generative AI
Without generative AI¹ Overall technical automation potential,
comparison in midpoint scenarios,
% in the United States in 2023
Share of US
employment, %
13
22
9
22
24
Bachelor’s degree
Associate’s degree
Some college
High school diploma or equivalent
No high school degree 9

“hollowing out of the middle,” and our previous models have suggested that work automation
would likely have the biggest midterm impact on lower-middle-income quintiles.
However, generative AI’s impact is likely to most transform the work of higher-wage
knowledge workers because of advances in the technical automation potential of their
activities, which were previously considered to be relatively immune from automation
(Exhibit 13).
Exhibit 13
Generative AI could have the biggest impact on activities in high-wage jobs;
previously, automation’s impact was highest in lower-middle-income quintiles.
McKinsey & Company
Automation adoption per wage quintile, % in 2030, midpoint scenario
United States Japan Germany France
China India Mexico South Africa
Without generative AI¹ With generative AI
81–100 61–80 41–60 21–40 0–20
Largest increase in automation
adoption from generative AI
Largest automation adoption
without generative AI
Wage quintiles Higher earners Lower earners
0
10
20
30
40
0
10
20
30
40
43

Generative AI Potential

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Generative AI Potential

Similar to Generative AI Potential (20)

More from Kapil Khandelwal (KK)

More from Kapil Khandelwal (KK) (20)

Recently uploaded

Recently uploaded (20)

Generative AI Potential