Michael Chui | San Francisco
James Manyika | San Francisco
Mehdi Miremadi | Chicago
Nicolaus Henke | London
Rita Chung | Silicon Valley
Pieter Nel | New York
Sankalp Malhotra | New York
DISCUSSION PAPER
APRIL 2018
NOTES FROM
THE AI FRONTIER
INSIGHTS FROM
HUNDREDS OF
USE CASES
2
McKinsey Global Institute

Copyright McKinsey & Company 2018
Since its founding in 1990, the McKinsey Global Institute (MGI) has sought to develop a deeper
understanding of the evolving global economy. As the business and economics research arm of
McKinsey & Company, MGI aims to provide leaders in the commercial, public, and social sectors with
the facts and insights on which to base management and policy decisions.
MGI research combines the disciplines of economics and management, employing the analytical tools
of economics with the insights of business leaders. Our "micro-to-macro" methodology examines
microeconomic industry trends to better understand the broad macroeconomic forces affecting
business strategy and public policy. MGI's in-depth reports have covered more than 20 countries and
30 industries. Current research focuses on six themes: productivity and growth, natural resources, labor
markets, the evolution of global financial markets, the economic impact of technology and innovation,
and urbanization. Recent reports have assessed the digital economy, the impact of AI and automation
on employment, income inequality, the productivity puzzle, the economic benefits of tackling gender
inequality, a new era of global competition, Chinese innovation, and digital and financial globalization.
MGI is led by three McKinsey & Company senior partners: Jacques Bughin, Jonathan Woetzel, and
James Manyika, who also serves as the chairman of MGI. Michael Chui, Susan Lund, Anu Madgavkar,
Jan Mischke, Sree Ramaswamy, and Jaana Remes are MGI partners, and Mekala Krishnan and
Jeongmin Seong are MGI senior fellows.
Project teams are led by the MGI partners and a group of senior fellows, and include consultants from
McKinsey offices around the world. These teams draw on McKinsey's global network of partners and
industry and management experts. Advice and input to MGI research are provided by the MGI Council,
members of which are also involved in MGI's research. MGI Council members are drawn from around
the world and from various sectors and include Andrs Cadena, Sandrine Devillard, Richard Dobbs,
Tarek Elmasry, Katy George, Rajat Gupta, Eric Hazan, Eric Labaye, Acha Leke, Scott Nyquist,
Gary Pinkus, Sven Smit, Oliver Tonby, and Eckart Windhagen. In addition, leading economists, including
Nobel laureates, act as research advisers to MGI research.
The partners of McKinsey fund MGI's research; it is not commissioned by any business, government,
or other institution. For further information about MGI and to download reports, please visit
www.mckinsey.com/mgi.
MCKINSEY ANALYTICS
McKinsey Analytics helps clients achieve better performance through data. We work together with
clients to build analytics-driven organizations, helping them develop the strategies, operations, and
capabilities to derive rapid and sustained impact from analytics. Over the past five years, we have
worked with more than 2,000 clients across every industry and business function. McKinsey Analytics
is led globally by Nicolaus Henke and Noshir Kaka, together with an executive committee comprised
of 40 McKinsey senior partners representing all regions and practices. Today, McKinsey Analytics
brings together more than 1,900 advanced analytics and AI experts and spans more than 125 domains
(industry- and function-specific teams with people, data, and tools focused on unique applications of
analytics). McKinsey Analytics includes several acquired companies such as QuantumBlack, a leading
advanced analytics firm that McKinsey acquired in 2015.
Learn more at www.mckinsey.com/business-functions/mckinsey-analytics/our-insights.
IN BRIEF
NOTES FROM THE AI FRONTIER:
INSIGHTS FROM HUNDREDS OF USE CASES
For this discussion paper, part of our ongoing research into evolving technologies and their
effect on business, economies, and society, we mapped traditional analytics and newer
"deep learning" techniques and the problems they can solve to more than 400 specific
use cases in companies and organizations. Drawing on MGI research and the applied
experience with artificial intelligence (AI) of McKinsey Analytics, we assess both the practical
applications and the economic potential of advanced AI techniques across industries and
business functions. We continue to study these AI techniques and additional use cases. For
now, here are our key findings:
AI, which for the purposes of this paper we characterize as "deep learning" techniques
using artificial neural networks, can be used to solve a variety of problems. Techniques
that address classification, estimation, and clustering problems are currently the most
widely applicable in the use cases we have identified, reflecting the problems whose
solutions drive value across the range of sectors.
The greatest potential for AI we have found is to create value in use cases in which more
established analytical techniques such as regression and classification techniques
can already be used, but where neural network techniques could provide higher
performance or generate additional insights and applications. This is true for 69 percent
of the AI use cases identified in our study. In only 16 percent of use cases did we find a
"greenfield" AI solution that was applicable where other analytics methods would not be
effective.
Because of the wide applicability of AI across the economy, the types of use cases with
the greatest value potential vary by sector. These variations primarily result from the
relative importance of different drivers of value within each sector. They are also affected
by the availability of data, its suitability for available techniques, and the applicability of
various techniques and algorithmic solutions. In consumer-facing industries such as
retail, for example, marketing and sales is the area with the most value. In industries
such as advanced manufacturing, in which operational performance drives corporate
performance, the greatest potential is in supply chain, logistics, and manufacturing.
The deep learning techniques on which we focused feed forward neural networks,
recurrent neural networks, and convolutional neural networksaccount for about
40 percent of the annual value potentially created by all analytics techniques. These
three techniques together can potentially enable the creation of between $3.5 trillion and
$5.8 trillion in value annually. Within industries, that is the equivalent of 1 to 9 percent of
2016 revenue.
Capturing the potential impact of these techniques requires solving multiple problems.
Technical limitations include the need for a large volume and variety of often labeled
training data, although continued advances are already helping address these. Tougher
perhaps may be the readiness and capability challenges for some organizations.
Societal concern and regulation, for example about privacy and use of personal data,
can also constrain AI use in banking, insurance, health care, and pharmaceutical and
medical products, as well as in the public and social sectors, if these issues are not
properly addressed.
The scale of the potential economic and societal impact creates an imperative for all
the participantsAI innovators, AI-using companies and policy-makersto ensure
a vibrant AI environment that can effectively and safely capture the economic and
societal benefits.
1
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
INTRODUCTION
Artificial intelligence (AI) stands out as a transformational
technology of our digital age. Questions about what it is, what
it can already doand what it has the potential to become
cut across technology, psychology, politics, economics,
science fiction, law, and ethics. AI is the subject of countless
discussions and articles, from treatises about technical
advances to tabloid headlines about its effects. Even as the
debate continues, the technologies underpinning AI continue
to move forward, enabling applications from facial recognition
in smartphones to consumer apps that use AI algorithms to
detect diabetes and hypertension with increasing accuracy.1
Indeed, while much of the public discussion of AI focuses on
science fiction-like AI realization such as robots, the number
of less-noticed practical applications for AI throughout the
economy is growing apace and permeating our lives.
This discussion paper seeks to contribute to the body
of knowledge about AI by mapping AI techniques to the
types of problems they can help solve and then mapping
these problem types to more than 400 practical use cases
and applications in businesses across 19 industries, from
aerospace and defense to travel and the public sector, and
nine business functions ranging from marketing and sales and supply-chain management
to product development and human resources.2 Drawing on a wide variety of public
and proprietary data sources, including the experiences of our McKinsey & Company
colleagues, we also assess the potential economic value of the latest generations of AI
technologies. The AI techniques we focus on are deep learning techniques based on
artificial neural networks, which we see as generating as much as 40 percent of the total
potential value that all analytics techniques could provide.
Our findings highlight the substantial potential of applying deep learning techniques to
use cases across the economy; these techniques can provide an incremental lift beyond
that from more traditional analytics techniques. We identify the industries and business
functions in which there is value to be captured, and we estimate how large that value
could be globally. For all the potential, much work needs to be done to overcome a range
of limitations and obstacles to AI application. We conclude with a brief discussion of these
obstacles and of future opportunities as the technologies continue their advance. Ultimately,
the value of AI is not to be found in the models themselves, but in organizations' abilities to
harness them. Business leaders will need to prioritize and make careful choices about how,
when, and where to deploy them.
This paper is part of our continuing research into analytics, automation, and AI technologies,
and their effect on business, the economy, and society.3 It is not intended to serve as a
comprehensive guide to deploying AI; for example, we identify but do not elaborate on
issues of data strategy, data engineering, governance, or change management and culture
1 Geoffrey H. Tison et al., "Cardiovascular risk stratification using off-the-shelf wearables and a multi-mask deep
learning algorithm," Circulation, volume 136, supplement 1, November 14, 2017.
2 We do not identify the companies by name or country, for reasons of client confidentiality.
3 Previous McKinsey Global Institute reports on these issues include The age of analytics: Competing in a data-
driven world, December 2016; A future that works: Automation, employment and productivity, January 2017;
and Artificial intelligence: The next digital frontier? June 2017. See a list of our related research at the end of
this paper.
What's inside
Introduction
Page 1
1. Mapping AI techniques
to problem types
Page 2
2. Insights from
use cases
Page 7
3. Sizing the potential
value of AI
Page 17
4. The road to impact
and value
Page 26
Acknowledgments
Page 31
2
McKinsey Global Institute
1. Mapping AI techniques to problem types
that are vital for companies seeking to capture value from AI and analytics.4 The use cases
we examined are not exhaustive; indeed, we continue to identify and examine others, and
we may update our findings in due course. Nonetheless, we believe that this research can
make a useful contribution to our understanding of what AI can and can't (yet) do, and
how much value could be derived from its use. It is important to highlight that, even as we
see economic potential in the use of AI techniques, the use of data must always take into
account concerns including data security, privacy, and potential issues of bias, issues we
have addressed elsewhere.5
1. MAPPING AI TECHNIQUES TO PROBLEM TYPES
As artificial intelligence technologies advance, so does the definition of which techniques
constitute AI (see Box 1, "Deep learning's origins and pioneers").6 For the purposes of this
paper, we use AI as shorthand specifically to refer to deep learning techniques that use
artificial neural networks. In this section, we define a range of AI and advanced analytics
techniques as well as key problem types to which these techniques can be applied.
NEURAL NETWORKS AND OTHER MACHINE LEARNING TECHNIQUES
We looked at the value potential of a range of analytics techniques. The focus of our
research was on methods using artificial neural networks for deep learning, which we
collectively refer to as AI in this paper, understanding that in different times and contexts,
other techniques can and have been included in AI. We also examined other machine
learning techniques and traditional analytics techniques (Exhibit 1). We focused on specific
potential applications of AI in business and the public sector (sometimes described
as "artificial narrow AI") rather than the longer-term possibility of an "artificial general
intelligence" that could potentially perform any intellectual task a human being is capable of.
4 See Jacques Bughin, Brian McCarthy, and Michael Chui, "A survey of 3,000 executives reveals how
businesses succeed with AI," Harvard Business Review, August 28, 2017.
5 Michael Chui, James Manyika, and Mehdi Miremadi, "What AI can and can't do (yet) for your business,"
McKinsey Quarterly, January 2018.
6 For a detailed look at AI techniques, see An executive's guide to AI, McKinsey Analytics, January 2018.
https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/an-executives-guide-to-ai
Exhibit 1
Techniques
Artificial intelligence, machine learning, and other analytics techniques that we examined for this research
SOURCE: McKinsey Global Institute analysis
Deep learning neural networks (e.g., feed
forward neural networks, CNNs, RNNs, GANs)
Reinforcement learning
Transfer learning
Advanced techniques
Traditional techniques
Linear classifiers (e.g., Fisher's
linear discriminant, SVM)
Monte Carlo
methods
Clustering (e.g., k-means,
tree based, db scan)
Regression Analysis (e.g.,
linear, logistic, lasso)
Markov process
(e.g., Markov chain)
Statistical inference (e.g.,
Bayesian inference, ANOVA)
Dimensionality reduction (e.g., PCA, tSNE)
Decision tree learning
Ensemble learning (e.g., random
forest, gradient boosting)
Instance based (e.g., KNN)
Naive Bayes classifier
Descriptive statistics
(e.g., confidence interval)
Likelihood to be used in
AI applications
Less
More
Considered AI for our research
3
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
Box 1: Deep learning's origins and pioneers
1 Warren McCulloch and Walter Pitts, "A logical calculus of the ideas immanent in nervous activity," Bulletin of
Mathematical Biophysics, volume 5, 1943.
2 Andrew Goldstein, "Bernard Widrow oral history," IEEE Global History Network, 1997.
3 Frank Rosenblatt, "The Perceptron: A probabilistic model for information storage and organization in the brain,"
Psychological review, volume 65, number 6, 1958.
4 Marvin Minsky and Seymour A. Papert, Perceptrons: An introduction to computational geometry, MIT Press,
January 1969.
5 David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams, "Learning representations by back-propagating
errors," Nature, volume 323, October 1986; for a discussion of Linnainmaa's role see Juergen Schmidhuber, Who
invented backpropagation?, Blog post http://people.idsia.ch/~juergen/who-invented-backpropagation.html,
2014.
6 Yann LeCun, Patrick Haffner, Leon Botton, and Yoshua Bengio, Object recognition with gradient-based learning,
Proceedings of the IEEE, November 1998.
7 John Hopfield, Neural networkds and physical systems with emergent collective computational abilities, PNAS,
April 1982.
8 Sepp Hochreiter and Juergen Schmidhuber, "Long short-term memory," Neural Computation, volume 9, number
8, December 1997.
9 Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton, ImageNet classification with deep convolutional neural
networks, NIPS 12 proceedings of the 25th International Conference on Neural Information Processing Systems,
December 2012.
10 Jeffrey Dean et al., Large scale distributed deep networks, NIPS 2012.
11 Richard S. Sutton and Andrew G. Barto, Reinforcement learning: An introduction, MIT Press, 1998.
12 Ian J. Goodfellow, Generative adversarial networks, ArXiv, June 2014.
It is too early to write a full history of deep learningand some of the details are contestedbut
we can already trace an admittedly incomplete outline of its origins and identify some of the
pioneers. They include Warren McCulloch and Walter Pitts, who as early as 1943 proposed
an artificial neuron, a computational model of the "nerve net" in the brain.1 Bernard Widrow
and Ted Hoff at Stanford University, developed a neural network application by reducing
noise in phone lines in the late 1950s.2 Around the same time, Frank Rosenblatt, an American
psychologist, introduced the idea of a device called the Perceptron, which mimicked the
neural structure of the brain and showed an ability to learn.3 MIT's Marvin Minsky and Seymour
Papert then put a damper on this research in their 1969 book "Perceptrons", by showing
mathematically that the Perceptron could only perform very basic tasks.4 Their book also
discussed the difficulty of training multi-layer neural networks. In 1986, Geoffrey Hinton at the
University of Toronto, along with colleagues David Rumelhart and Ronald Williams, solved this
training problem with the publication of a now famous back propagation training algorithm
although some practitioners point to a Finnish mathematician, Seppo Linnainmaa, as having
invented back propagation already in the 1960s.5 Yann LeCun at NYU pioneered the use
of neural networks on image recognition tasks and his 1998 paper defined the concept of
convolutional neural networks, which mimic the human visual cortex.6 In parallel, John Hopfield
popularized the "Hopfield" network which was the first recurrent neural network.7 This was
subsequently expanded upon by Jurgen Schmidhuber and Sepp Hochreiter in 1997 with
the introduction of the long short-term memory (LSTM), greatly improving the efficiency and
practicality of recurrent neural networks.8 Hinton and two of his students in 2012 highlighted
the power of deep learning when they obtained significant results in the well-known ImageNet
competition, based on a dataset collated by Fei-Fei Li and others.9 At the same time, Jeffrey
Dean and Andrew Ng were doing breakthrough work on large scale image recognition at
Google Brain.10 Deep learning also enhanced the existing field of reinforcement learning, led
by researchers such as Richard Sutton, leading to the game-playing successes of systems
developed by DeepMind.11 In 2014, Ian Goodfellow published his paper on generative
adversarial networks, which along with reinforcement learning has become the focus of much
of the recent research in the field.12 Continuing advances in AI capabilities have led to Stanford
University's One Hundred Year Study on Artificial Intelligence, founded by Eric Horvitz, building
on the long-standing research he and his colleagues have led at Microsoft Research. We have
benefited from the input and guidance of many of these pioneers in our research over the past
few years.
4
McKinsey Global Institute
1. Mapping AI techniques to problem types
Neural networks are a subset of machine learning techniques. Essentially, they are AI
systems based on simulating connected "neural units," loosely modeling the way that
neurons interact in the brain. Computational models inspired by neural connections have
been studied since the 1940s and have returned to prominence as computer processing
power has increased and large training data sets have been used to successfully analyze
input data such as images, video, and speech. AI practitioners refer to these techniques
as "deep learning," since neural networks have many ("deep") layers of simulated
interconnected neurons. Before deep learning, neural networks often had only three to five
layers and dozens of neurons; deep learning networks can have seven to ten or more layers,
with simulated neurons numbering into the millions.
In this paper, we analyzed the applications and value of three neural network techniques:
Feed forward neural networks. One of the most common types of artificial neural
network. In this architecture, information moves in only one direction, forward, from the
input layer, through the "hidden" layers, to the output layer. There are no loops in the
network. The first single-neuron network was proposed in 1958 by AI pioneer Frank
Rosenblatt. While the idea is not new, advances in computing power, training algorithms,
and available data led to higher levels of performance than previously possible.
Recurrent neural networks (RNNs). Artificial neural networks whose connections
between neurons include loops, well-suited for processing sequences of inputs, which
makes them highly effective in a wide range of applications, from handwriting, to texts,
to speech recognition. In November 2016, Oxford University researchers reported that
a system based on recurrent neural networks (and convolutional neural networks) had
achieved 95 percent accuracy in reading lips, outperforming experienced human lip
readers, who tested at 52 percent accuracy.
Convolutional neural networks (CNNs). Artificial neural networks in which the
connections between neural layers are inspired by the organization of the animal visual
cortex, the portion of the brain that processes images, well suited for visual perception
tasks.
We estimated the potential of those three deep neural network techniques to create value,
as well as other machine learning techniques such as tree-based ensemble learning,
classifiers, and clustering, and traditional analytics such as dimensionality reduction and
regression.
For our use cases, we also considered two other techniquesgenerative adversarial
networks (GANs) and reinforcement learningbut did not include them in our potential value
assessment of AI, since they remain nascent techniques that are not yet widely applied in
business contexts. However, as we note in the concluding section of this paper, they may
have considerable relevance in the future.
Generative adversarial networks (GANs). These usually use two neural networks
contesting each other in a zero-sum game framework (thus "adversarial"). GANs can
learn to mimic various distributions of data (for example text, speech, and images) and
are therefore valuable in generating test datasets when these are not readily available.
Reinforcement learning. This is a subfield of machine learning in which systems are
trained by receiving virtual "rewards" or "punishments," essentially learning by trial and
error. Google DeepMind has used reinforcement learning to develop systems that can
play games, including video games and board games such as Go, better than human
champions.
5
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
PROBLEM TYPES AND THE ANALYTIC TECHNIQUES THAT CAN BE APPLIED
TO SOLVE THEM
In a business setting, those analytic techniques can be applied to solve real-life problems.
For this research, we created a taxonomy of high-level problem types, characterized by the
inputs, outputs, and purpose of each. A corresponding set of analytic techniques can be
applied to solve these problems. These problem types include:
Classification. Based on a set of training data, categorize new inputs as belonging to
one of a set of categories. An example of classification is identifying whether an image
contains a specific type of object, such as a truck or a car, or a product of acceptable
quality coming from a manufacturing line.
Continuous estimation. Based on a set of training data, estimate the next numeric
value in a sequence. This type of problem is sometimes described as "prediction,"
particularly when it is applied to time series data. One example of continuous estimation
is forecasting the sales demand for a product, based on a set of input data such as
previous sales figures, consumer sentiment, and weather. Another example is predicting
the price of real estate, such as a building, using data describing the property combined
with photos of it.
Clustering. These problems require a system to create a set of categories, for which
individual data instances have a set of common or similar characteristics. An example
of clustering is creating a set of consumer segments based on data about individual
consumers, including demographics, preferences, and buyer behavior.
All other optimization. These problems require a system to generate a set of outputs
that optimize outcomes for a specific objective function (some of the other problem
types can be considered types of optimization, so we describe these as "all other"
optimization). Generating a route for a vehicle that creates the optimum combination of
time and fuel use is an example of optimization.
Anomaly detection. Given a training set of data, determine whether specific inputs are
out of the ordinary. For instance, a system could be trained on a set of historical vibration
data associated with the performance of an operating piece of machinery, and then
determine whether a new vibration reading suggests that the machine is not operating
normally. Note that anomaly detection can be considered a subcategory of classification.
Ranking. Ranking algorithms are used most often in information retrieval problems
in which the results of a query or request needs to be ordered by some criterion.
Recommendation systems suggesting next product to buy use these types of
algorithms as a final step, sorting suggestions by relevance, before presenting the results
to the user.
Recommendations. These systems provide recommendations, based on a set of
training data. A common example of recommendations are systems that suggest the
"next product to buy" for a customer, based on the buying patterns of similar individuals,
and the observed behavior of the specific person.
Data generation. These problems require a system to generate appropriately novel
data based on training data. For instance, a music composition system might be used to
generate new pieces of music in a particular style, after having been trained on pieces of
music in that style.
6
McKinsey Global Institute
1. Mapping AI techniques to problem types
Exhibit 2 illustrates the relative total value of these problem types across our database of use
cases, along with some of the sample analytics techniques that can be used to solve each
problem type. The most prevalent problem types are classification, continuous estimation,
and clustering, suggesting that meeting the requirements and developing the capabilities
in associated techniques could have the widest benefit. Some of the problem types that
rank lower can be viewed as subcategories of other problem typesfor example, anomaly
detection is a special case of classification, while recommendations can be considered
a type of optimization problemand thus their associated capabilities could be even
more relevant.
Exhibit 2
Problem types Sample techniques
% total AI value potential that could be unlocked by problem
types as essential vs. relevant to use cases
Classification
CNNs, logistic regression
Continuous
estimation
Feed forward neural networks, linear
regression
Clustering
K-means, affinity propagation
All other
optimization
Genetic algorithms
Anomaly
detection
One-class support vector machines,
k-nearest neighbors, neural networks
Ranking
Ranking support vector machines,
neural networks
Recommender
systems
Collaborative filtering
Data
generation
Generative adversarial networks
(GANs), hidden Markov models
44
37
16
17
19
9
14
29
29
39
21
6
8
7
7
15
1
0
72
17
66
24
37
55
Problem types and sample techniques
Relevant
Essential
SOURCE: McKinsey Global Institute analysis
NOTE: Sample techniques include traditional analytical techniques, machine learning, and the deep learning techniques we describe in this paper as AI.
Numbers may not sum due to rounding.
7
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
2. INSIGHTS FROM USE CASES
We collated and analyzed more than 400 use cases across 19 industries and nine business
functions for this research (see Box 2, "Our AI use cases"). They provided considerable
insight into the areas within specific sectors where deep neural networks can potentially
create the most value, the incremental performance lift that these neural networks can
generate compared with more traditional analytics, and the voracious data requirementsin
terms of volume, variety, and velocitythat must be met for this potential to be realized.
Examples of where AI can be used to improve the performance of existing use
cases include:
Predictive maintenance: the power of machine learning to detect anomalies. Some
existing predictive maintenance systems have analyzed time series data from Internet
of Things (IoT) sensors, such as those monitoring temperature or vibration, in order to
detect anomalies or make forecasts on the remaining useful life of components. Deep
learning's capacity to analyze very large amounts of high dimensional data can take this
to a new level. By layering in additional data, such as audio and image data, from other
sensorsincluding relatively cheap ones such as microphones and camerasneural
networks can enhance and possibly replace more traditional methods. AI's ability to
predict failures and allow planned interventions can be used to reduce downtime and
operating costs while improving production yield. In some industry examples we studied,
Box 2. Our AI use cases
We define a "use case" as a targeted application of
digital technologies to a specific business challenge,
with a measurable outcome. We recognize that use
cases can be described at different levels of granularity.
The use cases we analyzed for this paper correspond
to descriptions of specific business challenges
which practitioners in the industries and sectors we
studied acknowledged as meaningful. For example,
recommending the "next product to buy" for e-commerce
in the retail industry was considered a use case, whereas
"marketing and sales" was not sufficiently granular to
be considered a use case, even though it is a relevant
business function. We also collated similar use cases
across sectors into "domains" (see related section below).
For this research we built a library of use cases across
the economy that is as comprehensive as possible. The
cases that we identified come from a variety of sources,
including thousands of engagements by McKinsey
Analytics with clients around the globe. The data
incorporate findings from real-life examples of companies
and public-sector organizations using a range of analytics
techniques, both AI and more traditional approaches, as
well as the potential application of these techniques in
situations similar to those in which they have already been
successfully deployed. For example, where we found a
use case in one sector, such as a pricing promotion use
case in travel, we looked to see if it could be applied in
other sectors, for example in retail. Where possible, we
identified and analyzed multiple instances of individual
use cases.
For each use case, we estimated the annual value
potential of applying AI and other analytics across the
entire economy. This value potential could be captured
by companies and organizations themselves, in the form
of increased profits, or by their customers, in the form of
lower prices or higher quality. For use cases that involve
increasing revenue, such as those in sales and marketing,
we estimated the economy-wide value potential in terms
of the increased potential productivity of sales and
marketing expenditures, assuming that overall annual
spend in the economy is fixed in a given year (rather than
estimating the impact of higher revenues, which would
have assumed that overall spending would increase).
Our estimates are based on the structure of the global
economy in 2016. We did not estimate the value potential
of creating entirely new product or service categories,
such as autonomous driving.
Our library of use cases, while extensive, is not
exhaustive, and thus it may overstate or understate the
potential for certain sectors. We realize it may also contain
some biases based on the business profile of clients. Our
library of examples is a living one and we will continue
refining and adding to it.
8
McKinsey Global Institute
2. Insights from use cases
using remote on-board diagnostics to anticipate the need for service could theoretically
generate value of 1 to 2 percent of total sales. In a case involving cargo aircraft, AI can
extend the life of the plane beyond what is possible using traditional analytic techniques
by combining plane model data, maintenance history, IoT sensor data such as anomaly
detection on engine vibration data, and images and video of engine condition.
AI-driven logistics optimization can reduce costs through real-time forecasts
and behavioral coaching. Application of AI techniques such as continuous estimation
to logistics can add substantial value across many sectors. AI can optimize routing
of delivery traffic, thereby improving fuel efficiency and reducing delivery times. One
European trucking company has reduced fuel costs by 15 percent, for example. By
using sensors that monitor both vehicle performance and driver behavior, drivers
receive real-time coaching, including when to speed up or slow down, optimizing fuel
consumption and reducing maintenance costs. In another example, an airline uses AI to
predict congestion and weather-related problems in order to avoid costly cancellations.
For an airline with 100,000 flights per day, a 1 percent reduction in cancellations can
make a material difference
AI can be a valuable tool for customer service management and personalized
marketing. Improved speech recognition in call center management and call routing
by applying AI techniques allow a more seamless experience for customersand
more efficient processing. The capabilities go beyond words alone. For example, deep
learning analysis of audio allows systems to assess customers' emotional tone; if a
customer is responding negatively to an automated system, the call can be rerouted to
human operators and managers. In other areas of marketing and sales, AI techniques
can also have a significant impact. "Next product to buy" recommendations that target
individual customersas companies such as Amazon and Netflix have successfully
implementedcan lead to a substantial increase in the rate of sales conversions.
Individual pricing and promotion also has a broad application across sectors. For
example, in auto insurance, premiums can be customized based on data about driving
patterns and distances driven. In one use case, a travel company that micro-segmented
customers built a 360-degree customer view and offered additional services, such as
hotels and airlines, using a recommender system algorithm trained on product and
customer data. This led to a 10 to 15 percent increase in ancillary revenue. In retail, AI
can use SKU-performance data to optimize weekly and thematic product promotions,
including giving daily promotional recommendations.
MAPPING ANALYTICS TECHNIQUES TO USE CASES
For each use case, we catalogued the specific analytical techniques that could be applied,
including traditional analytics techniques, and various forms of machine learning, including
deep learning. We found that the applicability of different techniques varied across sectors
and business functions. Exhibits 3 and 4 are heatmaps that show the extent to which
applicable techniques can be used by industry and function, based on our library of use
cases. A select set of traditional techniques, including clustering, regression, and tree-
based models, are broadly applicable across many industries and functions. Similarly, the
more advanced AI techniques, such as RNNs, CNNs, and feed forward neural networks, are
relevant in many of the same industries and functions. The variations in the heat maps relate
to the different types of problems to be solved in the use cases, and the most applicable
techniques with which to do that.
9
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
Exhibit 3
Heat map: Technique relevance to industries
SOURCE: McKinsey Global Institute analysis
Focus of report
Traditional analytics techniques
Reinforcement learningFeed forward networksRecurrent neural networksConvolutional neural networksGenerative adversarial networksTree-based ensemble learningDimensionality reductionClassifiersClusteringRegression analysisStatistical inferenceMonte CarloMarkov processesOther optimizationAdvanced electronics/
semiconductors
Aerospace and
defense
Agriculture
Automotive and
assembly
Banking
Basic materials
Chemicals
Consumer packaged
goods
Health-care systems
and services
High tech
Insurance
Media and
entertainment
Oil and gas
Pharmaceuticals and
medical products
Public and social
sector
Retail
Telecommunications
Transport and logistics
Travel
Low
High
Number of use cases
10
McKinsey Global Institute
2. Insights from use cases
Use cases are often associated with multiple problem types: in fact, solving an average
of three problem types is either required for or applicable to a use case. Truck route
optimization provides one example. While an optimization solution is required to generate
most of the value by ensuring that vehicles take the most efficient routes, point estimation
could add incremental value. For example, this could be a model that predicts how long a
traffic light will stay red. With this knowledge, the route optimizer can determine whether to
suggest speeding up or slowing down to minimize vehicle stop time, which can be costly
from a fuel consumption standpoint. Thus, we catalogued problem types that are of primary
importance, and others that are relevant, for individual use cases.
The techniques with the broadest scope of applicability across industries and functions
include traditional analytic techniques, such as regression, tree-based ensemble learning,
classifiers, clustering, and other forms of statistical inference. That said, the neural network-
based techniques that we identify with the current generation of AI. AI also demonstrates
wide potential applicability across sectors and functions. However, their usage is not
yet widespread, partly because of the relative immaturity of the technology and the
organizational challenges of deploying these techniques. Among business functions, these
techniques are for now mostly to be found in marketing and sales and in supply-chain
management and manufacturing. In particular, feed forward neural networks feature as the
main technique deployed in these two functions.
Exhibit 4
Heat map: Technique relevance to functions
SOURCE: McKinsey Global Institute analysis
Low
High
Focus of report
Traditional analytics techniques
Reinforcement learningFeed forward networksRecurrent neural networksConvolutional neural networksGenerative adversarial networksTree-based ensemble learningDimensionality reductionClassifiersClusteringRegression analysisStatistical inferenceMonte CarloMarkov processesOther optimizationFinance and IT
Human resources
Marketing and sales
Other operations
Product development
Risk
Service operations
Strategy and corporate
finance
Supply-chain manage-
ment and manufacturing
Number of use cases
11
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
We have also collected individual use cases into broader "domains" that address closely
related problems. For example, the predictive maintenance domain aggregates use cases
from across multiple sectors and types of machinery in which the use of analytics and AI can
help to identify when repairs can be made to prevent a piece of equipment from breaking
down (Exhibit 5).
Overall, a deep understanding of use cases and how they are associated with particular
problem types, analytical techniques, and data types can help guide organizations
regarding where to invest in the technical capabilities and data that will provide the
greatest impact.
Exhibit 5
Use case domains
Structured/
semi-
structured
Time
series
Text
Audio
Video
Image
Analytics-driven accounting and IT
Analytics-driven hiring and retention
Channel management
Churn reduction
Customer acquisition/lead generation
Customer service management
Fraud and debt analytics
Inventory and parts optimization
Logistics network and warehouse optimization
Marketing budget allocation
Next product to buy/individualized offering
Predictive maintenance
Predictive service/intervention
Pricing and promotion
Procurement and spend analytics
Product development cycle optimization
Product feature optimization
Risk modeling
Sales and demand forecasting
Smart capital expenditures
Task automation
Workforce productivity and efficiency
Yield optimization
Use case mapping to data types
SOURCE: McKinsey Global Institute analysis
Low
High
Number of use cases
12
McKinsey Global Institute
2. Insights from use cases
TWO-THIRDS OF THE OPPORTUNITIES TO USE AI ARE IN IMPROVING THE
PERFORMANCE OF EXISTING ANALYTICS USE CASES
In 69 percent of the use cases we studied, deep neural networks can be used to improve
performance beyond that provided by other analytic techniques. Cases in which only
neural networks can be used, which we refer to here as "greenfield" cases, constituted just
16 percent of the total. For the remaining 15 percent, artificial neural networks provided
limited additional performance over other analytics techniques, among other reasons
because of data limitations that made these cases unsuitable for deep learning.
Greenfield AI solutions are prevalent in business areas such as customer service
management, as well as among some industries in which the data are rich and voluminous
and at times integrate human reactions. A key differentiator that often underpins higher AI
value potential is the possibility of applying large amounts of audio, video, image, and text
data to these problems. Among industries, we found many greenfield use cases in health
care, in particular. Some of these cases involve disease diagnosis and improved care, and
rely on rich data sets incorporating image and video inputs, including from MRIs.
On average, our use cases suggest that modern deep learning AI techniques have the
potential to provide a boost in value above and beyond traditional analytics techniques
ranging from 30 percent to 128 percent, depending on industry (Exhibit 6).
In many of our use cases, however, traditional analytics and machine learning techniques
continue to underpin a large percentage of the value creation potential in industries including
insurance, pharmaceuticals and medical products, and telecommunications, with the
potential of AI limited in certain contexts. In part this is due to the way data are used by those
industries and to regulatory issues, as we discuss later in this paper.
DATA REQUIREMENTS FOR DEEP LEARNING ARE SUBSTANTIALLY GREATER
THAN FOR OTHER ANALYTICS, IN TERMS OF BOTH VOLUME AND VARIETY
Making effective use of neural networks in most applications requires large labeled training
data sets alongside access to sufficient computing infrastructure. As the size of the training
data set increases, the performance of traditional techniques tends to plateau in many
cases. However, the performance of advanced AI techniques using deep neural networks,
configured and trained effectively, tends to increase. Furthermore, these deep learning
techniques are particularly powerful in extracting patterns from complex, multi-dimensional
data types such as images, video, and audio or speech. The data will need to be collected in
a way that addresses society's concerns about issues of privacy.
Data volume is essential for neural networks to achieve a high level of accuracy
in training algorithms
Deep learning methods require thousands of data records for models to become relatively
good at classification tasks and, in some cases, millions for them to perform at the level
of humans. By one estimate, a supervised deep-learning algorithm will generally achieve
acceptable performance with around 5,000 labeled examples per category and will match
or exceed human level performance when trained with a data set containing at least
10 million labeled examples.7 In some cases in which advanced analytics is currently used,
so much data are availablemillions or even billions of rows per data setthat AI usage is
the most appropriate technique. However, if a threshold of data volume is not reached, AI
may not add value to traditional analytics techniques.
7
Ian Goodfellow, Yoshua Bengio, and Aaron Courville, Deep learning, MIT Press, 2016.
13
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
Exhibit 6
In more than two-thirds of our use cases, AI can improve performance beyond that provided by other analytics
techniques
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding.
Potential incremental value from AI
over other analytics techniques
Breakdown of use cases by
applicable techniques
69
16
15
Full value can be
captured using
non-AI techniques
AI necessary to
capture value
("greenfield")
AI can improve
performance over
that provided by
other analytics
techniques
30
36
38
39
44
44
44
50
55
55
56
57
67
79
85
85
87
89
128
Aerospace and defense
Advanced electronics/
semiconductors
Average = 62
Oil and gas
Transport and logistics
Retail
Automotive and assembly
High tech
Public and social sector
Insurance
Pharmaceuticals and
medical products
Telecommunications
Healthcare systems
and services
Agriculture
Consumer packaged goods
Media and entertainment
Chemicals
Basic materials
Travel
Banking
%
14
McKinsey Global Institute
2. Insights from use cases
These massive data sets can be difficult to obtain or create for many business use cases,
and labeling remains a challenge. For example, teaching an autonomous vehicle to navigate
urban traffic requires enormous image data sets in which all critical objects (other vehicles of
all types, traffic signs, pedestrians, road markings, and so on) are labeled under all weather
and lighting conditions. Most current AI models are trained through "supervised learning",
which requires humans to label and categorize the underlying data. However promising
new techniques are emerging to overcome these data bottlenecks, such as reinforcement
learning, generative adversarial networks, transfer learning, and one-shot learning.8 Unlike
reinforcement learning, in which systems learn to perform tasks by trial and error, one-shot
learning allows a trained AI model to learn about a subject based on a small number of real-
world demonstrations or examplesand sometimes just one.
Nonetheless, our research shows that almost three-quarters of the impact from advanced
analytics is tied to use cases requiring millions of labeled data examples. This means that
organizations will have to adopt and implement strategies that enable them to collect,
integrate, and process data at scale. They will need to ensure that as much of their data
as possible is captured electronically, typically by progressing in their overall digitization
journeys. Even with large datasets, they will have to guard against "overfitting," in which
a model too tightly matches the "noisy" or random features of the training set, resulting
in a corresponding lack of accuracy in future performance, and against "underfitting," in
which the model fails to capture all of the relevant features. Linking data across customer
segments and channels and, where possible, to production data, rather than allowing
different sets of data to languish in silos, is especially important to create value. To achieve
this linkage, companies will need to create meta-data models, and manage both the internal
challenges as well as the regulatory risks of sharing the data across a range of business
functions.
Realizing AI's full potential requires a diverse range of data types including
images, video, and audio
Neural AI techniques excel at analyzing image, video, and audio data types because of their
complex, multi-dimensional nature, known by practitioners as "high dimensionality." Just a
single training instance can have many different features, often requiring multiple levels of
analysis. These systems can thus require orders of magnitude more training data than other
machine learning systems with lower dimensionality data.
However, while images, video, and audio data are particularly well-suited for use with
modern deep learning techniques, even more value can be extracted from mining insights
from traditional structured and time series data types. Exhibit 7 highlights the range of value
from our use cases based on the types of data used.
In the past few years, advances in deep learning have pushed AI-driven image classification
tasks beyond human performance in some cases. An image of a face, for example, consists
of thousands of pixels that can be combined to form eyes, ears, and other features. When
arranged in the right way, they form the face of a specific person. Neural networks are good
at dealing with high dimensionality, as multiple layers in a network can learn to represent
the many different features present in the data. Thus, for facial recognition, the first layer
in the network could focus on raw pixels, the next on edges and lines, another on generic
facial features, and the final layer might identify the face. Unlike previous generations of AI,
which often required human expertise to do "feature engineering," these neural network
techniques are often able to learn to represent these features in their simulated neural
networks as part of the training process.
8 Michael Chui, James Manyika, and Mehdi Miremadi, "What AI can and can't do (yet) for your business,"
McKinsey Quarterly, January 2018; Matthew Hutson, "The future of AI depends on a huge workforce of
human teachers," Bloomberg Businessweek, September 7, 2017.
15
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
An example of the progress that has been made can be found in the results of the annual
ImageNet Large Scale Visual Recognition Challenge (Exhibit 8).9 Since 2010 machine
learning researchers have competed by submitting algorithms for detecting objects within
a public database of 14 million labeled images. The increase in accuracy observed in
2012 is widely considered the harbinger of the deep learning revolution; for the first time,
a deep neural net was used to address the problem and showed dramatic improvement
over previous efforts. The accuracy of the best-performing algorithms (all now using deep
learning) now exceeds "human level performance"the accuracy level expected of a human
performing the same task.
9 Olga Russakovsky et al., "ImageNet Large Scale Visual Recognition Challenge," International Journal of
Computer Vision, volume 115, issue 3, December 2015.
Exhibit 7
% of total value potential
Range of potential AI value impact by data type
SOURCE: McKinsey Global Institute analysis
1626
Time series
Structured
Text
Image
Audio
5083
5594
Video
1730
713
2542
Data type
Range
Exhibit 8
The ability of AI systems to recognize objects has improved markedly to the point where the best systems now
exceed human performance
15
12
2017
16
90
80
13
11
2010
70
0
75
85
100
95
14
Accuracy
%
Human performance
Best AI system
SOURCE: ImageNet Large Scale Visual Recognition Challenge; McKinsey Global Institute analysis
16
McKinsey Global Institute
2. Insights from use cases
These increases in performance using deep learning techniques enabled many of the
consumer products we already take for granted, including services such as Siri, Alexa, and
Cortana. Improved image-processing technology underpins some of the capabilities that
are being tested in self-driving cars today.
Ongoing data acquisition for retraining AI systems is necessary; one out of
three use cases requires model refreshes at least monthly and sometimes daily
An analysis of our use cases shows that, along with issues around the volume and variety of
data, velocity is also a requirement: AI techniques require models to be retrained to match
potential changing conditions, so the training data must be refreshed frequently. In one-third
of the cases, the model needs to be refreshed at least monthly, and almost one in four cases
requires a daily refresh; this is especially true in marketing and sales and in supply-chain
management and manufacturing (Exhibit 9).
Exhibit 9
For about one-third of use cases, the models require frequent updating: three-quarters of those cases require
monthly refreshes, while nearly one-quarter are at least weekly
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding.
Share of use cases
34
At least monthly
refreshes
Less frequent
refreshes
66
77
23
At least
weekly
At least
monthly
Frequency of refresh required
17
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
3. SIZING THE POTENTIAL VALUE OF AI
An analysis of the value derived from our use cases suggests that AI can generate
considerable value. We estimate that the AI techniques we cite in this reportfeed forward
neural networks and convolutional neural networkstogether have the potential to create
between $3.5 trillion and $5.8 trillion in value annually across nine business functions in 19
industries. The value as measured by percentage of industry revenue varies significantly
among industries, depending on the specific applicable use cases, the availability of
abundant and complex data, and regulatory and other constraints. While we found patterns
in the potential of AI to create value within industries and functions, every company will need
to look at the details of its business to estimate the opportunities to create value enabled
by AI.
AI COULD POTENTIALLY CREATE $3.5 TRILLION TO $5.8 TRILLION IN ANNUAL
VALUE IN THE GLOBAL ECONOMY
We estimated a range of annual value to the global economy for both AI and other analytics
techniques, based on the value we observed being created in current use cases and the
potential value in projected future ones. The total annual value potential of AI alone across 19
industries and nine business functions in the global economy came to between $3.5 trillion
and $5.8 trillion. This constitutes about 40 percent of the overall $9.5 trillion to $15.4 trillion
annual impact that could potentially be enabled by all analytical techniques (Exhibit 10).
Per industry, we estimate that AI's potential value amounts to between 1 and 9 percent of
2016 revenue. Even the industry with the smallest potential value at stake, aerospace and
defense (less than $50 billion) could enable the annual creation of value that is equivalent to
the GDP of Lebanon.
These figures are not forecasts for a particular period in time, but they are indicative of the
considerable potential for the global economy that advanced analytics represents. The
percentage of this potential that individual organizations, sectors, and functions can achieve
will be the product of competitive and market dynamics, as well as of many choices and
decisionsand indeed business model choicesby organizations and others. They include
the developers of technology, policy makers who set the overall context, and customers
who choose what to purchase. Some of this value will be captured in a variety of ways, for
example it may result in more valued products and services, revenue growth, cost savings,
or indeed consumer surplus. While the aggregate numbers may appear modest, in some
use cases the advancements amount to radical transformation.
For AI to realize a substantial portion of the potential value we have estimated will require
companies to deploy these techniques comprehensively in areas where they can most
effectively harness their ability to make the complexity of data an advantage. Take the
travel industry (by which we refer to all aspects of commercial passenger travel, from travel
agents and airlines to hotels and online providers), in which the potential AI impact can
more than double what is achievable using traditional analytic methods, amounting to
between 7 and almost 12 percent of total revenue for the industry. To achieve that potential
will require AI deployment in top-line revenue-related marketing and sales use cases, and
bottom-line operations use cases, where in both cases the data are rich and the value of
each incremental percentage point performance increase is significant. As an example of
one single AI use case of the many that would be required for this industry to achieve its
full potential, Hawaii's state tourism authority, working with a major online travel company,
uses facial recognition software to monitor travelers' expressions through their computer
webcams and deliver personalized offers.10
10 The Hawai'i tourism authority and Expedia Media Solutions use custom-built facial recognition software to
create personalized travel marketing campaign, Expedia Group, press release, September 26, 2016.
18
McKinsey Global Institute
3. Sizing the potential value of AI
Exhibit 10
AI has the potential to create annual value across sectors totaling $3.5 trillion to $5.8 trillion, or 40 percent of the
overall potential impact from all analytics techniques
200
25
55
50
60
500
100
45
40
0
300
700
600
0
35
400
30
20
Media and
entertainment
Chemicals
Advanced electronics/
semiconductors
Automotive
and assembly
Transport
and logistics
Travel
Aerospace
and defense
Al impact as % of total impact derived from analytics
Banking
Oil and gas
Agriculture
Telecommunications
Pharmaceuticals
and medical products
Basic materials
Health-care systems
and services
High tech
Al impact
$ billion
Public and
social sector
Consumer
packaged goods
Insurance
Retail
0.10.1
Aerospace and defense
Pharmaceuticals and medical products
Chemicals
0.20.2
Advanced electronics/semiconductors
Oil and gas
Insurance
0.20.3
0.10.2
0.20.3
Banking
0.10.2
Telecommunications
Media and entertainment
0.10.2
Agriculture
0.10.3
High tech
0.20.3
Basic materials
0.10.2
<0.1T
Travel
Transport and logistics
Automotive and assembly
0.20.3
0.30.5
0.20.5
0.20.3
0.40.5
Health-care systems and services
0.40.8
Consumer packaged goods
Retail
0.30.4
Public and social sector
0.30.4
Aggregate dollar impact ($ trillion)
Impact as % of industry revenues
3.35.3
1.02.3
4.26.1
2.54.9
2.96.9
2.43.7
1.81.9
2.93.7
2.96.3
2.64.0
1.11.4
3.27.1
5.710.2
1.63.1
1.8-3.2
4.96.4
2.55.2
3.25.7
7.211.6
The potential value of AI by sector
SOURCE: McKinsey Global Institute analysis
NOTE: Artificial Intelligence here includes neural networks only. Numbers may not sum due to rounding.
19
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
By contrast, as a percentage of overall revenues, the potential annual value of the AI
use cases we identified in telecommunications is between about 3 and 6 percent of
industry revenue (still amounting to over $100 billion in potential annual impact). While
telecommunications operators have a large volume of customer data, much of that data
is in structured forms for which firms can just as easily use more conventional analytics
and machine learning techniques to leverage that data. The public sector also has a large
volume of data, and a range of data types and use cases that make it a ripe area for AI
applications. However, regulations and requirements for data privacy and interpretability,
particularly in developed markets, present significant challenges to its use, and thus
constrain the value potential.
Even with those caveats, industries for which we have estimated lower AI impact in terms of
percentages of overall revenue compared with other sectors, such as aerospace, defense,
and the public sector, nonetheless have the potential to create billions or even hundreds
of billions of dollars of value from its deployment where appropriate. A key benefit of this
exercise across industries is to provide executives and leaders in each industry with a
perspective for where the largest potential opportunities lie. This will help them set priorities
as well as identifying where they can benefit from AI.
THE BIGGEST VALUE OPPORTUNITIES FOR AI ARE IN MARKETING AND SALES
AND IN SUPPLY-CHAIN MANAGEMENT AND MANUFACTURING
From the use cases we have examined, we find that the greatest potential value impact from
using AI are both in top-line-oriented functions, such as marketing and sales, and in bottom-
line-oriented operational functions, including supply-chain management and manufacturing
(Exhibit 11). At a company level, every firm will need to examine its mix of functions to find the
most attractive opportunities to use AI, and determine where it makes most sense to invest
in AI deployment.
From the analysis of our use cases we can see broad applications of AI in marketing
and sales across industries including consumer packaged goods, banking, retail,
telecommunications, high tech, travel, insurance, and media and entertainment. Indeed,
marketing and sales and supply-chain management together constitute some of the biggest
areas of opportunity for AI (Exhibit 12). As noted earlier, AI is a powerful tool for personalizing
product recommendations including through analyzing aggregated user data to understand
individual customer preferences. Companies will nonetheless still need to think through
business models that involve data use in such cases.
Consumer industries such as retail and high tech will tend to see more potential from
marketing and sales AI applications because frequent and digital interactions between
business and customers generate larger data sets for AI techniques to tap into.
E-commerce platforms, in particular, stand to benefit. This is because of the ease with
which these platforms collect customer information, such as click data or time spent on a
web page, and can then customize promotions, prices, and products for each customer
dynamically and in real time. For their part, brick-and-mortar retailers can implement AI
applications to improve product assortment and inventory management per store, and to
optimize their supply chains end-to-end. Indeed, they can take advantage of the Internet
of Things to generate data usable by AI techniques to both improve the performance of
their supply chains and apply some of the top-line innovations from the online world to the
offline world, for example, by viewing dwell time in front of a physical display as analogous to
spending more time viewing web or mobile content.11
11 See The Internet of Things: Mapping the value beyond the hype, McKinsey Global Institute, June 2015.
20
McKinsey Global Institute
3. Sizing the potential value of AI
Exhibit 11
AI's potential impact is greatest in marketing and sales and supply-chain management and manufacturing,
based on our use cases
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding.
Value potential
$ trillion
Marketing
and sales
3.36.0
1.42.6
Supply-chain
management and manufacturing
3.65.6
1.22.0
0.50.9
Risk
Finance
and IT
0.2
0.2
0.1
0.2
0.1
HR
0.6
0.2
Service
operations
0.3
0.1
0.3
Product
development
<0.1
Strategy and
corporate
finance
Value potential
By all analytics (darker color)
$9.5 trillion15.4 trillion
By AI (lighter color)
$3.5 trillion5.8 trillion
0.91.3
0.20.4
Other
operations
21
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
In our use cases, for example, we found that using real-time data to predict hyper-
regional demand trends can increase sales by 0.25 percent to 0.75 percent, with margin
improvements from lower waste and spoilage amounting to as much as half of one
percentage point of sales. The impact can be considerably larger in pharmaceutical and
medical products, in which predicting hyper-regional product demand and relevant health
trends to inform inventory levels and reduce spoilage has the potential to raise sales by 5 to
10 percent.
Exhibit 12
Marketing and sales and supply-chain management and manufacturing are among the functions where AI can
create the most incremental value
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding.
Highest potential impact business problems per functional area
Impact size comparison by chart area)
$ trillion
Yield
optimization
0.30.6
Analytics-
driven HR
0.1
Smart
capex
<0.1
Product
feature
optimiza-
tion
<0.1
Product
develop-
ment cycle
0.1
Predictive
service/
interven-
tion
0.2
Risk
0.1
Fraud
and
debt
analy-
tics
0.1
Analytics-
driven
accounting
and IT
0.10.2
Task
automation
0.10.2
Workforce
productivity
and efficiency
0.10.2
0.1
0.1
Procurement and spend
analytics
0.10.2
Inventory and parts
optimization
0.10.2
Predictive
maintenance
0.50.7
0.1
Channel
manage-
ment
0.10.2
Churn
reduction
0.10.2
Customer
acquisition/
lead generation
0.10.3
Price and
promotion
0.30.5
Next product
to buy
(NPTB)
individual-
ized offering
0.30.5
Customer service
management
0.40.8
Marketing and sales
Supply-chain management
and manufacturing
Other
Marketing
budget
allocation
Logistics network
and warehouse
optimization
Sales and
demand
forecast
22
McKinsey Global Institute
3. Sizing the potential value of AI
AI's ability to conduct preventive maintenance and field force scheduling, as well as
optimizing production and assembly processes, means that it also has considerable
application possibilities and value potential across sectors including advanced electronics
and semiconductors, automotive and assembly, chemicals, basic materials, transportation
and logistics, oil and gas, pharmaceuticals and medical products, aerospace and defense,
agriculture, and consumer packaged goods. In advanced electronics and semiconductors,
for example, harnessing data to adjust production and supply-chain operations can
minimize spending on utilities and raw materials, cutting overall production costs by 5 to
10 percent in our use cases.
THE OPPORTUNITIES TO CREATE VALUE WITH AI CORRESPOND TO THE
VALUE DRIVERS WITHIN INDUSTRIES
Identifying the opportunity for AI deployment depends on a range of factors that are specific
to individual sectors and, within those sectors, to different types of businesses. Given the
wide range of applicability of AI techniques, broadly speaking, if you want to know where
AI can create the most value, you need to follow the money. For industries in which the
main drivers of value are related to marketing and sales, including many consumer-facing
industries, that is where the greatest value from deploying AI can be found. However, for
industries in which the key driver of value is operational excellence, such as in advanced
manufacturing and oil and gas, functions such as supply chain and manufacturing are those
in which AI can create the most value.
It is instructive to compare industry sectors in terms of where AI has the greatest value
potential. A detailed breakdown by sector is available on our interactive chart online. Here,
by way of illustration, we take a snapshot of three sectorsretail, consumer packaged
goods, and bankingwhere we have seen AI's impact (Exhibits 1315). In retail, marketing
and sales is the area with the most significant potential value from AI, and within that
function, pricing and promotion and customer service management are the main value
areas. Our use cases show that using customer data to personalize promotions, for
example, including tailoring individual offers every day, can lead to a 1 to 2 percent increase
in incremental sales for brick-and-mortar retailers alone. In packaged goods, supply-
chain management is the key function that could benefit from AI deployment. Among the
examples in our use cases, we see how forecasting based on underlying causal drivers of
demand rather than prior outcomes can improve forecasting accuracy by 10 to 20 percent,
which translates into a potential 5 percent reduction in inventory costs and revenue
increases of 2 to 3 percent. In banking, particularly retail banking, AI has significant value
potential in marketing and sales, much as it does in retail. Although the banking industry has
been among the fastest to adopt AI and capture the value in risk management, the benefits
could be even greater for other industries. For example, AI is helping insurers price risk
more accurately, pharmaceutical companies are using AI to reduce risk in clinical trials, and
mining firms have deployed the technologies to anticipate disruptions to production.
23
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
Exhibit 13
In retail, AI has the most potential impact in pricing and promotion and other marketing and sales areas
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding. Not to scale.
Retail examples
$ trillion
Marketing and sales
0.30.5
~0.1
Finance
and IT
<0.05
HR
<0.05
Other
operations
0.10.2
<0.05
<0.05
Supply-chain
management and
manufacturing
Product development
Strategy and
corporate finance
Pricing and
promotion
0.10.2
Customer
service
management
~0.1
Next product
to buy
<0.05
Customer
acquisition
and
generation
<0.1
<0.05
Task automation
0.10.2
<0.05
<0.05
<0.05
Inventory and parts
optimization
<0.1
<0.05
<0.05
<0.05
Logistics network
Marketing
budget
allocation
Workplace
productivity
and efficiency
Analytics-driven
accounting
and IT
Analytics-driven
hiring and
retention
Product
feature
optimization
Strategy and
corporate
finance
24
McKinsey Global Institute
3. Sizing the potential value of AI
As our library of use cases evolves, we expect to see significant value in some areas that is
not fully captured in these figures. One example is risk in banking. We believe that this could
be one of the biggest areas of impact for deep learning using neural networks. For example,
incorporating AI into underwriting models could allow banks to underwrite entirely new
types of customers, such as the unbanked or semi-banked, and capture significant value
through improved fraud detection. The application of neural network techniques to these
new use cases in risk will need to be further tested before we can size the impact.
Exhibit 14
In consumer packaged goods, AI's greatest potential is in supply-chain management and manufacturing, including
predictive maintenance
Packaged goods examples
$ trillion
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding. Not to scale.
Marketing
and sales
~0.1
Supply-chain
management and
manufacturing
0.20.3
Finance
and IT
<0.05
HR
<0.05
Other
opera-
tions
<0.05
Yield optimization
<0.1
Predictive maintenance
~0.1
Inventory and parts
optimization
~0.1
<0.05
<0.05
<0.05
Channel
manage-
ment
<0.05
0.05
<0.05
<0.05
<0.05
<0.05
<0.05
Product development
cycle optimization
Analytics-driven
hiring and retention
Sales and
demand
forecasting
Logistics network
and warehouse
optimization
Pricing and
promotion
Marketing budget
allocation
Analytics-driven
accounting and IT
Other
operations
Procurement and
spend analytics
Product
develop-
ment
0.05
25
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
Exhibit 15
In banking, marketing and sales and risk are among the areas with the most potential value from AI
SOURCE: McKinsey Global Institute analysis
NOTE: Numbers may not sum due to rounding. Not to scale.
Banking example
$ trillion
Workforce
productivity
and efficiency
<0.05
Analytics-
driven hiring
and retention
<0.05
Analytics-
driven finance
and IT
<0.05
Fraud and debt analytics
00.1
Customer acquisition/
lead generation
<0.05
Next product to buy
(NPTB)
<0.05
Pricing and promotion
<0.05
Churn reduction
0.05
Customer service
management
00.1
Channel
management
0.1
Marketing and sales
0.10.2
Risk
00.1
Finance
and IT
<0.05
HR
<0.05
Other
operations
<0.05
26
McKinsey Global Institute
4. The road to impact and value
4. THE ROAD TO IMPACT AND VALUE
Artificial intelligence is attracting growing amounts of corporate investment, and as the
technologies develop, the potential value that can be unlocked is likely to grow. So far,
however, only a few pioneering firms have adopted AI at scale. Prior research suggests
that even among AI-aware firms, only about 20 percent are using one or more of the
technologies in a core business process or at scale.12 For all their promise, AI technologies
have plenty of limitations that will need to be overcome, including not just data-related issues
but also regulatory obstacles, and social and user acceptance. Yet the potential value to be
harnessed provides a clear incentive for technology developers, companies, policy-makers,
and users to try to tackle these issues.
WHILE AI IS PROMISING, ITS USE STILL FACES LIMITATIONS AND
CHALLENGES
As discussed, factors that could limit AI use include the requirements around the volume,
type, and labeling of data. Other limitations are also significant, for now, although the
evolving technologies themselves are starting to provide some solutions.
Limitations include the need for massive data sets, difficulties in explaining
results, generalizing learning, and potential bias in data and algorithms
Among the limitations we have identified, five stand out.13 First is the challenge of labeling
training data, which often must be done manually and is necessary for supervised learning.
Ironically, machine learning often requires large amounts of manual effort; in supervised
learning, the set of machine learning techniques that is most often used, machines are
taught; they don't learn "by themselves." Promising new techniques are emerging to
address this challenge, such as reinforcement learning (discussed earlier) and in-stream
supervision, in which data can be labeled in the course of natural usage.14 Second is the
difficulty of obtaining data sets that are sufficiently large and comprehensive to be used for
training; for many business use cases, creating or obtaining such massive data sets can be
difficultfor example, limited clinical-trial data to predict health-care treatment outcomes
more accurately. Third is the difficulty of explaining in human terms results from large and
complex models: why was a certain decision reached? Product certifications in health care,
as well as in the automotive, chemicals, and aerospace industries, for example, can be an
obstacle; among other constraints, regulators often want rules and choice criteria to be
clearly explainable. Some nascent approaches to increasing model transparency, including
local-interpretable-model-agnostic explanations (LIME), may help resolve this explanation
challenge in many cases.15 Fourth is the generalizability of learning: AI models continue to
have difficulties in carrying their experiences from one set of circumstances to another. That
means companies must commit resources to train new models even for use cases that are
similar to previous ones. Transfer learningin which an AI model is trained to accomplish
a certain task and then quickly applies that learning to a similar but distinct activity, is one
promising response to this challenge.16
The fifth limitation concerns the risk of bias in data and algorithms. Unlike the other
limitations listed, which may eventually be resolved by technical advances, this issue of bias
touches on concerns that are more social in nature and which could require broader steps
to resolve, such as understanding how the processes used to collect training data can
12 Ibid. McKinsey Global institute, Artificial intelligence, June 2017.
13 Ibid. Michael Chui, James Manyika, and Mehdi Miremadi, "What AI can and can't do (yet) for your business,"
McKinsey Quarterly, January 2018.
14 Eric Horvitz, "Machine learning, reasoning, and intelligence in daily life: Directions and challenges,"
Proceedings of Artificial Intelligence Techniques for Ambient Intelligence, Hyderabad, India, January 2007.
15 LIME attempts to identify which parts of input data a trained model relies on most to make predictions.
16 For an early example application, see John Guttag, Eric Horvitz, and Jenna Wiens, "A study in transfer
learning: Leveraging data from multiple hospitals to enhance hospital-specific predictions," Journal of the
American Medical Informatics Association, volume 21, number 4, 2014.
27
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
influence the behavior of models they are used to train. In certain instances, when applied
incorrectly, AI models risk perpetuating existing social and cultural biases. For example,
unintended biases can be introduced when training data are not representative of the larger
population to which an AI model is applied. Thus, data collected from a city-wide video
camera network will likely show more instances of all kinds of signals, from crime to outdoor
advertising engagements, in areas where the density of deployed cameras is higher, which
might not be representative of the number of instances of these occurrences in a city as a
whole. Similarly, facial recognition models trained on a population of faces corresponding
to the demographics of artificial intelligence developers could struggle when applied to
populations with more diverse characteristics.17
Moreover, a recent report on the malicious use of AI highlights a range of security threats,
from sophisticated automation of hacking to hyper-personalized political disinformation
campaigns.18 Among the risks it lists are threats associated with privacy invasion and social
manipulation from the use of AI to automate tasks involved in surveillance, such as analyzing
mass-collected data, persuasion, including creating targeted propaganda, and deception,
for example with manipulated videos. Multiple research efforts are under way to identify best
practices and address such issues in academic, non-profit, and private-sector research.
Organizational challenges around technology, processes, and people can slow
or impede AI adoption
Organizations planning to adopt significant deep learning efforts will need to consider
a spectrum of options about how to do so. The range of options includes building a
complete in-house AI capability either gradually in an organic way or more rapidly through
acquisitions, outsourcing these capabilities, or leveraging AI-as-a-service offerings.
Given the importance of data, it is vital for organizations to develop strategies for the
creation and/or acquisition of training data. But the effective application of AI also requires
organizations to address other key data challenges, including establishing effective data
governance, defining ontologies, data engineering around the "pipes" from data sources,
managing models over time, building the data pipes from AI insights to either human or
machine actions, and managing regulatory constraints.
Given the significant computational requirements of deep learning, some organizations will
maintain their own data centers, because of regulations or security concerns, but the capital
expenditures could be considerable, particularly when using specialized hardware. Cloud
vendors offer another option.
Process can also become an impediment to successful adoption unless organizations
are digitally mature. On the technical side, organizations will have to develop robust data
maintenance and governance processes, and implement modern software disciplines such
as Agile and DevOps. Even more challenging, in terms of scale, is overcoming the "last mile"
problem of making sure the superior insights provided by AI are instantiated in the behavior
of the people and processes of an enterprise.
On the people front, much of the construction and optimization of deep neural networks
remains something of an art requiring real experts to deliver step-change performance
increases. Demand for these skills far outstrips supply at present; according to some
estimates fewer than 10,000 people have the skills necessary to tackle serious AI problems
17 Joy Buolamwini and Timnit Gebru. "Gender shades: Intersectional accuracy disparities in commercial
gender classification," Proceedings of Machine Learning Research, 2018. http://proceedings.mlr.press/v81/
buolamwini18a/buolamwini18a.pdf
18 Miles Brundage et al., The malicious use of artificial intelligence: Forecasting, prevention, and mitigation,
Future of Humanity Institute, February 2018.
28
McKinsey Global Institute
4. The road to impact and value
and competition for them is fierce amongst the tech giants.19 Companies wanting to build
their own AI solutions will need to consider whether they have the capacity to attract and
retain these specialized skills.
AI can seem an elusive business case for now
Where AI techniques and data are available and the value is clearly proven, organizations
can already pursue the opportunity. But in some areas, the techniques today may be
mature and the data available, but the cost and complexity of deploying AI may simply not
be worthwhile, given the value that could be generated. For example, an airline could use
facial recognition and other biometric scanning technology to streamline aircraft boarding,
but the value of doing so may not justify the cost and issues around privacy and personal
identification. A recent Stanford University study found that deep neural networks can make
highly accurate bond price predictions, but took hours to come up with the answer, whereas
other "simpler" techniques produced an answer that was only slightly less accurate but very
rapidjust four seconds.20 For a bond trader that timing difference is critical.
Similarly, we can see potential cases in which the data and the techniques are maturing, but
the value is not yet clear. For example, in mining, AI could potentially play a significant role in
providing ore body insights, thereby allowing for more efficient exploration, drilling, and mine
planning. Given the high capital expenditure costs usually involved in this sector, the benefits
could be very significant, but for now are unmeasurable. In other sectors, including banking,
opportunities that may exist are sometimes in areas closely guarded as competitive secrets.
In the most unpredictable scenario of all, either the data (both the types and volume) or the
techniques are simply too new and untested to know how much value they could unlock.
For example, in health care, if AI could build on the superhuman precision we are already
starting to see with X-ray analysis and broaden that to more accurate diagnoses and even
automated medical procedures, the economic value could be very significant. At the same
time, the complexities and costs of arriving at this frontier are also daunting. Among other
issues, it would require flawless technical execution and resolving issues of malpractice
insurance and other legal concerns.
Societal concerns and regulations can constrain AI use
Societal concerns and regulations can affect potential value capture from AI, including
in use cases related to personally identifiable information. This is particularly relevant at a
time of growing public debate about the use and commercialization of individual data on
some online platforms.21 Use and storage of personal information is especially sensitive
in sectors such as banking, health care, and pharmaceutical and medical products, as
well as in the public and social sector. Alongside these questions about privacy, issues of
fairness and equity may arise from bias in data, as well as concerns about transparency and
accountability in the use of massively complex algorithms. In addition to addressing these
issues, businesses and other users of data for AI will need to continue to evolve business
models related to data use in order to address societies' concerns. Furthermore, regulatory
requirements and restrictions can differ from country to country, as well from sector to
sector. In the European Union, for example, automated individual decision makingthat is,
algorithms that make decisions based on user-level predictorswill be shaped by the EU-
wide general data protection regulation taking effect in 2018, which provides for a right to an
explanation for some decisions made by machines. This could affect the insurance industry,
19 Cade Metz, "Tech giants are paying huge salaries for scarce AI talent," The New York Times, October 22,
2017.
20 "Neural networks face unexpected problems in analyzing financial data," MIT Technology Review, May 10,
2017.
21 Julia Angwin, Dragnet Nation: A quest for privacy, security, and freedom in a world of relentless surveillance,
Times Books, 2014; see also Anna Bernasek and D. T. Mongan, All you can pay: How companies use our
data to empty our wallets, Nation Books, 2015.
29
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
for example, in which underwriters could be required to explain how machines derived
their answers about whether to underwrite the risk or explain why the premium was set at a
certain level.
IMPLICATIONS FOR STAKEHOLDERS
As we have seen, it is a company's ability to execute against AI models, rather than the
models themselves, that creates value. In prior publications, we have detailed some of the
general implications for companies as they adopt AI and analytics.22 In this final section, we
sketch out some of the high-level implications of our study of AI use cases for providers of AI
technology, appliers of AI technology, and policy makers who set the context for both.
For AI technology provider companies
Many companies that develop or provide AI to others have considerable strength in the
technology itself and the data scientists needed to make it work, but they can lack a
deep understanding of end markets. This is a challenge, since our research has shown
that most of the potential impact of AI comes from improving the performance in existing
use casesin other words, the fundamental drivers of the businesses of their potential
customers. Furthermore, many of these companies are asking how they should prioritize
their resources, whether it is R&D, marketing and sales, or other functions to take advantage
of AI opportunities.
Understanding the value potential of AI across sectors and functions can help shape the
portfolios of these AI technology companies. That said, they shouldn't necessarily only
prioritize the areas of highest potential value. Instead, they can combine that data with
complementary analyses of the competitor landscape, of their own existing strengths
including technology and data strengths, their sector or function knowledge, and their
customer relationships, to shape their investment portfolios.
On the technical side, the mapping of problem types and techniques to sectors and
functions of potential value can guide a company with specific areas of expertise as to where
to focus. Similarly, for technology companies that have access to certain types of data, this
mapping can help guide them to where their access to data can provide them with the most
leverage, and toward data gaps needing to be filled. Finally, more innovation is needed to
continue to advance the frontier of AI and address some ongoing technical limitations.
For companies adopting AI to transform and power their own businesses
Many companies seeking to adopt AI in their operations have started machine learning
and AI experiments across their businessand are likely to be bombarded by technology
companies trying to sell them "AI solutions." Before launching more pilots or testing
solutions, it is useful to step back and take a holistic approach to the issue, moving to
create a prioritized portfolio of initiatives across the enterprise, including AI and the wider
analytic and digital techniques available. For a business leader to create an appropriate
portfolio, it is important to develop an understanding about which use cases and domains
have the potential to drive the most value for a company, as well as which AI and other
analytical techniques will need to be deployed to capture that value. This portfolio ought to
be informed not only by where the theoretical value can be captured, but by the question of
how the techniques can be deployed at scale across the enterprise. The question of how
analytical techniques are scaling is driven less by the techniques themselves and more by a
company's skills, capabilities, and data. Companies will need to consider efforts on the "first
mile," that is, how to acquire and organize data and efforts, as well as on the "last mile", or
how to integrate the output of AI models into work flows, ranging from clinical trial managers
and sales force managers to procurement officers. Previous MGI research suggests that
22 See for example, Artificial intelligence: The next digital frontier? McKinsey Global Institute, June 2017 and The
age of analytics: Competing in a data-driven world, McKinsey Global Institute, December 2016.
30
McKinsey Global Institute
4. The road to impact and value
AI leaders invest heavily in these first- and last-mile efforts.23 In addition, given growing user
and societal concerns, companies deploying AI will need to think through their safe and
responsible data use and the business models they employ that make use of the user or
customer data.
For policy makers
Policy makers will need to strike a balance between supporting the development of AI
technologies and managing any risks from bad actors, as well as irresponsible use of AI
techniques and the data they employ. They have an interest in supporting broad adoption,
since AI can lead to higher labor productivity, economic growth, and societal prosperity.
Tools to help them include public investments in research and development as well as
support for a variety of training programs, which can help nurture AI talent. On the issue
of data, governments can spur the development of training data directly through open
data initiatives. Opening up public-sector data can spur private-sector innovation. Setting
common data standards can also help.
AI is also raising new questions for policy makers to grapple with for which historical tools
and frameworks may not be adequate.24 Therefore, some policy innovations will likely
be needed to cope with these rapidly evolving technologies. But given the scale of the
beneficial impact on business, the economy, and society, the goal should not be to constrain
the adoption and application of AI, but rather to encourage its beneficial and safe use.

The several hundred use cases we examined underscore the value and performance
enhancement that adoption of AI technologies can bring, and provide some indication of
where AI can most usefully be deployed. To capture that value, CEOs and other C-suite
executives will need to ramp up and staff their analytics capabilities; as our use cases
show, AI can most often be adopted and create value where other analytics methods
and techniques are also creating value. For companies seeking to deploy AI, that means
the same basics will need to be in place, especially moving forward on digitizing their
enterprises. As we have seen, abundant volumes of rich data from images, audio, and
video, and large-scale text are the essential starting point and lifeblood of creating value
with AI. Above all, our analysis of use cases suggests that successful AI adoption will require
focus and setting of priorities. Its value is tremendousand looks set to become even more
so as the technologies themselves advance. Identifying where and how that value can be
captured looks likely to become one of the key business challenges of our era.
23 The age of analytics: Competing in a data-driven world, McKinsey Global Institute, December 2016.
24 Bank risk management is one example. See Ignacio Crespo, Pankaj Kumar, Peter Noteboom, and Marc
Taymans, The evolution of model risk management, McKinsey & Company, February 2017.
31
McKinsey Global Institute
Notes from the AI frontier: Insights from hundreds of use cases
ACKNOWLEDGMENTS
This discussion paper was produced as part of the McKinsey Global Institute's research on
the impact of technology on business and society, and specifically our on-going research
program on the future of work and the potential effects on the global economy of data and
analytics, automation, robotics, and artificial intelligence.
The research was led by Michael Chui, an MGI partner in San Francisco; James Manyika,
chairman and director of the McKinsey Global Institute and McKinsey senior partner
based in San Francisco; Mehdi Miremadi, a McKinsey partner based in Chicago;
and Nicolaus Henke, a McKinsey senior partner in London who co-leads McKinsey
Analytics and is chairman of Quantum Black, a McKinsey company. Martin Harrysson,
Martin McQuade, Roger Roberts, and Tamim Saleh helped guide the research. Rita Chung,
Ira Chadha, and Sankalp Malhotra headed the working team, which comprised Ali Akhtar,
Adib Ayay, and Pieter Nel.
We are grateful to colleagues within McKinsey who provided valuable advice and analytical
support: Luis Almeida, Ramnath Balasubramanian, Shubham Banerjee, Gaurav Batra,
Harold Bauer, Michele Bertoncello, Patrick Briest, Bede Broome, Ondrej Burkacky,
Mary Calam, Brian Carpizo, Michael Chen, Frank Coleman, Alex Cosmas, Reed Doucette,
Carlos Fernandes, David Fiacco, Kevin Goering, Ben Goodier, Taras Gorishnyy, Pete Grove,
Ludwig Hausmann, Michael van Hoey, Aaron Horowitz, Minha Hwang, Venkat Inumella,
Pallav Jain, Harold Janin, Mithun Kamat, Matthias Klasser, Rohit Kumar, Shannon Lijek,
Oskar Linqvist, Carl March, Brian McCarthy, Ryan McCullough, Doug McElhaney,
Andres Meza, Jordi Monso, Sumit Mundra, Florian Neuhaus, Kris Otis, Naveen Sastry,
Eric Schweikert, Jules Seeley, Richard Sellschop, Abdul Wahab Shaikh, Owen Stockdale,
Chris Thomas, Jeffrey Thompson, Richard Ward, Kate Whittington, Georg Winkler, and
Christoph Wollersheim.
We would like to thank Sandeep Gupta, Rajat Monga, and Martin Wicke of the Google Brain
team for their input on some of the technical issues. We have also benefited greatly from
the research of a range of leading practitioners and thinkers on AI and our dialogues with
them, including on its limitations and the technical advances to address them. They include
Jack Clark, strategy and communications director at OpenAI, Jeffrey Dean, lead of Google
Brain, Barbara J. Grosz, Higgins Professor of Natural Sciences at Harvard University,
Demis Hassabis, founder and CEO of DeepMind, Eric Horvitz, director of Microsoft
Research Labs, Kai-Fu Lee, CEO of Sinovation Ventures, Fei-Fei Li, director of Stanford
University's Artificial Intelligence Lab, and Andrew Ng, adjunct professor at Stanford
University. We also would like to thank and acknowledge the insights of Jacomo Corbo,
chief scientist, and other colleagues at Quantum Black, a McKinsey company.
This discussion paper was edited and produced by MGI editorial director Peter Gumbel,
editorial production manager Julie Philpot, and senior graphic designers Marisa Carder
and Patrick White, and graphic design specialist Margo Shimasaki. Rebeca Robboy, MGI
director of external communications, managed dissemination and publicity, while digital
editor Lauren Meling provided support for online and social media treatments.
This report contributes to MGI's mission to help business and policy leaders understand
the forces transforming the global economy, identify strategic locations, and prepare for the
next wave of growth. As with all MGI research, this work is independent and has not been
commissioned or sponsored in any way by any business, government, or other institution.
While we are grateful for all the input we have received, the report is ours, including any
errors. We welcome your comments on this research at MGI@mckinsey.com.
McKinsey Global Institute
April 2018
Copyright McKinsey & Company
www.mckinsey.com/mgi
@McKinsey_MGI
McKinseyGlobalInstitute