Organizational challenges

Organizational
challenges
Disruptive business
models
Enhanced decision
making
34 55 75
HIGHLIGHTS
IN COLLABORATION WITH
MCKINSEY ANALYTICS
DECEMBER 2016
THE AGE OF ANALYTICS:
COMPETING IN A
DATA-DRIVEN WORLD

Nicolaus Henke | London
Jacques Bughin | Brussels
Michael Chui | San Francisco
James Manyika | San Francisco
Tamim Saleh | London
Bill Wiseman | Taipei
Guru Sethupathy | Washington, DC
THE AGE OF ANALYTICS:
COMPETING IN A
DATA-DRIVEN WORLD
DECEMBER 2016
PREFACE
Five years ago, the McKinsey Global Institute (MGI) released Big data: The next frontier for
innovation, competition, and productivity
. In the years since, data science has continued to
make rapid advances, particularly on the frontiers of machine learning and deep learning.
Organizations now have troves of raw data combined with powerful and sophisticated
analytics tools to gain insights that can improve operational performance and create new
market opportunities. Most profoundly, their decisions no longer have to be made in the
dark or based on gut instinct; they can be based on evidence, experiments, and more
accurate forecasts.
As we take stock of the progress that has been made over the past five years, we see
that companies are placing big bets on data and analytics. But adapting to an era of
more data-driven decision making has not always proven to be a simple proposition for
people or organizations. Many are struggling to develop talent, business processes, and
organizational muscle to capture real value from analytics. This is becoming a matter of
urgency, since analytics prowess is increasingly the basis of industry competition, and
the leaders are staking out large advantages. Meanwhile, the technology itself is taking
major leaps forward—and the next generation of technologies promises to be even more
disruptive. Machine learning and deep learning capabilities have an enormous variety of
applications that stretch deep into sectors of the economy that have largely stayed on the
sidelines thus far.
This research is a collaboration between MGI and McKinsey Analytics, building on more
than five years of research on data and analytics as well as knowledge developed in work
with clients across industries. This research also draws on a large body of MGI research on
digital technology and its effects on productivity, growth, and competition. It aims to help
organizational leaders understand the potential impact of data and analytics, providing
greater clarity on what the technology can do and the opportunities at stake.
The research was led by Nicolaus Henke, global leader of McKinsey Analytics, based
in London; Jacques Bughin, an MGI director based in Brussels; Michael Chui, an MGI
partner based in San Francisco; James Manyika, an MGI director based in San Francisco;
Tamim Saleh, a senior partner of McKinsey based in London; and Bill Wiseman, a
senior partner of McKinsey based in Taipei. The project team, led by Guru Sethupathy
and Andrey Mironenko, included Ville-Pekka Backlund, Rachel Forman, Pete Mulligan,
Delwin Olivan, Dennis Schwedhelm, and Cory Turner. Lisa Renaud served as senior editor.
Sincere thanks go to our colleagues in operations, production, and external relations,
including Tim Beacom, Marisa Carder, Matt Cooke, Deadra Henderson, Richard Johnson,
Julie Philpot, Laura Proudlock, Rebeca Robboy, Stacey Schulte, Margo Shimasaki, and
Patrick White.
We are grateful to the McKinsey Analytics leaders who provided guidance across
the research, including Dilip Bhattacharjee, Alejandro Diaz, Mikael Hagstroem, and
Chris Wigley. In addition, this project benefited immensely from the many McKinsey
colleagues who shared their expertise and insights. Thanks go to Ali Arat, Matt Ariker,
Steven Aronowitz, Bill Aull, Sven Beiker, Michele Bertoncello, James Biggin-Lamming,
Yves Boussemart, Chad Bright, Chiara Brocchi, Bede Broome, Alex Brotschi, David Bueno,
Eric Buesing, Rune Bundgaard, Sarah Calkins, Ben Cheatham, Joy Chen, Sastry Chilukuri,

Brian Crandall, Zak Cutler, Seth Dalton, Severin Dennhardt, Alexander DiLeonardo,
Nicholas Donoghoe, Jonathan Dunn, Leeland Ekstrom, Mehdi El Ouali, Philipp Espel,
Matthias Evers, Robert Feldmann, David Frankel, Luke Gerdes, Greg Gilbert,
Taras Gorishnyy, Josh Gottlieb, Davide Grande, Daina Graybosch, Ferry Grijpink,
Wolfgang Günthner, Vineet Gupta, Markus Hammer, Ludwig Hausmann, Andras Havas,
Malte Hippe, Minha Hwang, Alain Imbert, Mirjana Jozic, Hussein Kalaoui, Matthias Kässer,
Joshua Katz, Sunil Kishore, Bjorn Kortner, Adi Kumar, Tom Latkovic, Daniel Läubli,
Jordan Levine, Nimal Manuel, J.R. Maxwell, Tim McGuire, Doug McElhaney,
Fareed Melhem, Phillipe Menu, Brian Milch, Channie Mize, Timo Möller, Stefan Nagel,
Deepali Narula, Derek Neilson, Florian Neuhaus, Dimitri Obolenski, Ivan Ostojic,
Miklos Radnai, Santiago Restrepo, Farhad Riahi, Stefan Rickert, Emir Roach,
Matthias Roggendorf, Marcus Roth, Tom Ruby, Alexandru Rus, Pasha Sarraf,
Whitney Schumacher, Jeongmin Seong, Sha Sha, Abdul Wahab Shaikh, Tatiana Sivaeva,
Michael Steinmann, Kunal Tanwar, Mike Thompson, Rob Turtle, Jonathan Usuka,
Vijay Vaidya, Sri Velamoor, Richard Ward, Khilony Westphely, Dan Williams, Simon Williams,
Eckart Windhagen, Martin Wrulich, Ziv Yaar, and Gordon Yu.
Our academic adviser was Martin Baily, Senior Fellow and Bernard L. Schwartz Chair in
Economic Policy Development at the Brookings Institution, who challenged our thinking and
provided valuable feedback and guidance. We also thank Steve Langdon and the Google
TensorFlow group for their helpful feedback on machine learning.
This report contributes to MGI’s mission to help business and policy leaders understand
the forces transforming the global economy and prepare for the next wave of growth.
As with all MGI research, this work is independent, reflects our own views, and has not
been commissioned by any business, government, or other institution. We welcome your
comments on the research at [email protected]
Jacques Bughin
Director, McKinsey Global Institute
Senior Partner, McKinsey & Company
Brussels
James Manyika
Director, McKinsey Global Institute
Senior Partner, McKinsey & Company
San Francisco
Jonathan Woetzel
Director, McKinsey Global Institute
Senior Partner, McKinsey & Company
Shanghai
December 2016

© Chombosan/Shutterstock
CONTENTS
HIGHLIGHTS
The demand for talent
Radical personalization in
health care
Machine learning and the
automation of work activities
38
66
87
In Brief
Page vi
Executive summary
Page 1
1. The data and analytics revolution gains momentum
Page 21
2. Opportunities still uncaptured
Page 29
3. Mapping value in data ecosystems
Page 43
4. Models of disruption fueled by data and analytics
Page 55
5. Deep learning: The coming wave
Page 81
Technical appendix
Page 95
Bibliography
Page 121
IN BRIEF
THE AGE OF ANALYTICS:
COMPETING IN A DATA-DRIVEN WORLD
Data and analytics capabilities have made a leap forward in recent years. The volume of available data has
grown exponentially, more sophisticated algorithms have been developed, and computational power and
storage have steadily improved. The convergence of these trends is fueling rapid technology advances
and business disruptions.
ƒ Most companies are capturing only a fraction of the potential value from data and analytics. Our 2011
report estimated this potential in five domains; revisiting them today shows a great deal of value still on
the table. The greatest progress has occurred in location-based services and in retail, both areas with
digital native competitors. In contrast, manufacturing, the public sector, and health care have captured
less than 30 percent of the potential value we highlighted five years ago. Further, new opportunities
have arisen since 2011, making the gap between the leaders and laggards even bigger.
ƒ The biggest barriers companies face in extracting value from data and analytics are organizational;
many struggle to incorporate data-driven insights into day-to-day business processes. Another
challenge is attracting and retaining the right talent—not only data scientists but business translators
who combine data savvy with industry and functional expertise.
ƒ Data and analytics are changing the basis of competition. Leading companies are using their
capabilities not only to improve their core operations but to launch entirely new business models. The
network effects of digital platforms are creating a winner-take-most dynamic in some markets.
ƒ Data is now a critical corporate asset. It comes from the web, billions of phones, sensors, payment
systems, cameras, and a huge array of other sources—and its value is tied to its ultimate use. While
data itself will become increasingly commoditized, value is likely to accrue to the owners of scarce
data, to players that aggregate data in unique ways, and especially to providers of valuable analytics.
ƒ Data and analytics underpin several disruptive models. Introducing new types of data sets
(“orthogonal data”) can disrupt industries, and massive data integration capabilities can break through
organizational and technological silos, enabling new insights and models. Hyperscale digital platforms
can match buyers and sellers in real time, transforming inefficient markets. Granular data can be used
to personalize products and services—and, most intriguingly, health care. New analytical techniques
can fuel discovery and innovation. Above all, data and analytics can enable faster and more evidencebased decision making.
ƒ Recent advances in machine learning can be used to solve a tremendous variety of problems—and
deep learning is pushing the boundaries even further. Systems enabled by machine learning can
provide customer service, manage logistics, analyze medical records, or even write news stories. The
value potential is everywhere, even in industries that have been slow to digitize. These technologies
could generate productivity gains and an improved quality of life—along with job losses and other
disruptions. Previous MGI research found that 45 percent of work activities could potentially be
automated by currently demonstrated technologies; machine learning can be an enabling technology
for the automation of 80 percent of those activities. Breakthroughs in natural language processing
could expand that impact even further.
Data and analytics are already shaking up multiple industries, and the effects will only become more
pronounced as adoption reaches critical mass. An even bigger wave of change is looming on the horizon
as deep learning reaches maturity, giving machines unprecedented capabilities to think, problem-solve,
and understand language. Organizations that are able to harness these capabilities effectively will be able
to create significant value and differentiate themselves, while others will find themselves increasingly at
a disadvantage.

Only a fraction of the value we envisioned in 2011 has been captured to date
Enhanced sensory
perception
Understanding
natural language
Recognizing
known patterns
Generating
natural language
Optimizing and
planning
The age of analytics:
Competing in a data-driven world
Data and analytics fuel 6 disruptive models that
change the nature of competition
Machine learning has broad applicability in many common work activities
Data
Data-driven
discovery and
innovation
99% 79% 76% 59% 33%
Generate Aggregate Analyze
Value
Generate Aggregate Analyze
Volume
Massive
data
integration
Hyperscale,
real-time
matching
Enhanced
decision
making
Radical
personalization
European Union
public sector
United States
health care
Manufacturing United States
retail
Location-based
data
Orthogonal
data sets
As data ecosystems evolve, value
will accrue to providers of analytics,
but some data generators and
aggregators will have unique value
Value share
Percent of work activities that require:
Volume of data and use cases per player
10–20% 10–20% 20–30% 30–40% 50–60%
viii McKinsey Global Institute
© B. Busco/Getty Images
EXECUTIVE SUMMARY
Back in 2011, the McKinsey Global Institute published a report highlighting the
transformational potential of big data.
1 Five years later, we remain convinced that this
potential has not been overhyped. In fact, we now believe that our 2011 analyses gave only a
partial view. The range of applications and opportunities has grown even larger today.
The convergence of several technology trends is accelerating progress. The volume of
data continues to double every three years as information pours in from digital platforms,
wireless sensors, and billions of mobile phones. Data storage capacity has increased, while
its cost has plummeted. Data scientists now have unprecedented computing power at their
disposal, and they are devising ever more sophisticated algorithms.
The companies at the forefront of these trends are using their capabilities to tackle business
problems with a whole new mindset. In some cases, they have introduced data-driven
business models that have taken entire industries by surprise. Digital natives have an
enormous advantage, and to keep up with them, incumbents need to apply data and
analytics to the fundamentals of their existing business while simultaneously shifting the
basis of competition. In an environment of increasing volatility, legacy organizations need
to have one eye on high-risk, high-reward moves of their own, whether that means entering
new markets or changing their business models. At the same time, they have to apply
analytics to improve their core operations. This may involve identifying new opportunities
on the revenue side, using analytics insights to streamline internal processes, and building
mechanisms for experimentation to enable continuous learning and feedback.
Organizations that pursue this two-part strategy will be ready to take advantage of
opportunities and thwart potential disruptors—and they have to assume that those
disruptors are right around the corner. Data and analytics have altered the dynamics in many
industries, and change will only accelerate as machine learning and deep learning develop
capabilities to think, problem-solve, and understand language. The potential uses of these
technologies are remarkably broad, even for sectors that have been slow to digitize. As we
enter a world of self-driving cars, personalized medicine, and intelligent robots, there will be
enormous new opportunities as well as significant risks—not only for individual companies
but for society as a whole.
MOST COMPANIES ARE CAPTURING ONLY A FRACTION OF THE POTENTIAL
VALUE OF DATA AND ANALYTICS
Turning a world full of data into a data-driven world is an idea that many companies have
found difficult to pull off in practice. Our 2011 report estimated the potential for big data and
analytics to create value in five specific domains. Revisiting them today shows both uneven
progress and a great deal of that value still on the table (Exhibit E1).
We see the greatest progress in location-based services and in US retail. In contrast,
adoption is lagging in manufacturing, the EU public sector, and US health care. Incentive
problems and regulatory issues pose additional barriers to adoption in the public sector
and health care. In several cases, incumbent stakeholders that would have the most to lose
from the kinds of changes data and analytics could enable also have a strong influence on
regulations, a factor that could hinder adoption.
1 Big data: The next frontier for innovation, competition, and productivity, McKinsey Global Institute, June 2011.
2 McKinsey Global Institute Executive summary
ƒ Location-based services: GPS-enabled smartphones have put mapping technology
in the pockets of billions of users. The markets for global positioning system navigation
devices and services, mobile phone location-based service applications, and geotargeted mobile advertising services have reached 50 to 60 percent of the value we
envisioned in 2011. End consumers are capturing the lion’s share of the benefits, mostly
through time and fuel savings as well as new types of mobile services. Beyond the value
we envisioned in 2011, there are growing opportunities for businesses to use geospatial
data to track assets, teams, and customers across dispersed locations in order to
generate new insights and improve efficiency.
ƒ US retail: Retailers can mine a trove of transaction-based and behavioral data from
their customers. Thin margins (especially in the grocery sector) and pressure from
industry-leading early adopters such as Amazon and Walmart have created strong
incentives to put that data to work in everything from cross-selling additional products to
reducing costs throughout the entire value chain. The US retail sector has realized 30 to
40 percent of the potential margin improvements and productivity growth we envisioned
in 2011, but again, a great deal of value has gone to consumers.
ƒ Manufacturing: Manufacturing industries have achieved only about 20 to 30 percent
of the potential value we estimated in 2011—and most has gone to a handful of industry
leaders. Within research and design, design-to-value applications have seen the
greatest uptick in adoption, particularly among carmakers. Some industry leaders have
developed digital models of the entire production process (“digital factories”). More
companies have integrated sensor data-driven operations analytics, often reducing
Exhibit E1
Potential impact: 2011 research
Value captured
% Major barriers
Location-based
data
ƒ $100 billion+ revenues for service providers
ƒ Up to $700 billion value to end users
ƒ Penetration of GPS-enabled
smartphones globally
US retail1 ƒ 60%+ increase in net margin
ƒ 0.5–1.0% annual productivity growth
ƒ Lack of analytical talent
ƒ Siloed data within companies
Manufacturing2 ƒ Up to 50% lower product development cost
ƒ Up to 25% lower operating cost
ƒ Up to 30% gross margin increase
ƒ Siloed data in legacy IT systems
ƒ Leadership skeptical of impact
EU public
sector
3
ƒ ~€250 billion value per year
ƒ ~0.5% annual productivity growth
ƒ Lack of analytical talent
ƒ Siloed data within different
agencies
US health care ƒ $300 billion value per year
ƒ ~0.7% annual productivity growth
ƒ Need to demonstrate clinical
utility to gain acceptance
ƒ Interoperability and data sharing
There has been uneven progress in capturing value from data and analytics
SOURCE: Expert interviews; McKinsey Global Institute analysis
1 Similar observations hold true for the EU retail sector.
2 Manufacturing levers divided by functional application.
3 Similar observations hold true for other high-income country governments.
50–
60
30–
40
20–
30
10–
20
10–
20
Future of decision making (big data)
ES
mc 1205
REPEATS in report
McKinsey Global Institute The age of analytics: Competing in a data-driven world 3
operating costs by 5 to 15 percent. After-sales servicing offers are beginning to be based
on real-time surveillance and predictive maintenance.
ƒ The EU public sector: Our 2011 report analyzed how the European Union’s public
sector could use data and analytics to make government services more efficient,
reduce fraud and errors in transfer payments, and improve tax collection, potentially
achieving some €250 billion worth of annual savings. But only about 10 to 20 percent of
this has materialized. Some agencies have moved more interactions online, and many
(particularly tax agencies) have introduced pre-filled forms. But across Europe and other
advanced economies, adoption and capabilities vary greatly. The complexity of existing
systems and the difficulty of attracting scarce analytics talent with public-sector salaries
have slowed progress. Despite this, we see even wider potential today for societies to
use analytics to make more evidence-based decisions in many aspects of government.
ƒ US health care: To date, only 10 to 20 percent of the opportunities we outlined in 2011
have been realized by the US health-care sector. A range of barriers—including a lack of
incentives, the difficulty of process and organizational changes, a shortage of technical
talent, data-sharing challenges, and regulations—have combined to limit adoption.
Within clinical operations, the biggest success has been the shift to electronic medical
records, although the vast stores of data they contain have not yet been fully mined.
While payers have been slow to capitalize on big data for accounting and pricing, a
growing industry now aggregates and synthesizes clinical records, and analytics have
taken on new importance in public health surveillance. Many pharmaceutical firms are
using analytics in R&D, particularly in streamlining clinical trials. While the health-care
sector continues to lag in adoption, there are enormous unrealized opportunities to
transform clinical care and deliver personalized medicine (a topic we will return to below).
LEGACY COMPANIES HAVE TO OVERCOME HURDLES TO ACCELERATE THEIR
ANALYTICS TRANSFORMATION
The relatively slow pace of progress in some of the domains described above points to the
fact that many companies that have begun to deploy data and analytics have not realized
the full value. Some have responded to competitive pressure by making large technology
investments but have failed to make the organizational changes needed to make the most
of them.
An effective transformation strategy can be broken down into several components
(Exhibit E2). The first step should be asking some fundamental questions to shape the
strategic vision: What will data and analytics be used for? How will the insights drive value?
How will the value be measured? The second element is building out the underlying data
architecture as well as data collection or generation capabilities. Many incumbents struggle
with switching from legacy data systems to a more nimble and flexible architecture to store
and harness big data. They may also need to digitize their operations more fully in order to
capture more data from their customer interactions, supply chains, equipment, and internal
processes. Looking at a wide variety of indicators that measure digitization, we see a striking
gap between leading firms and average firms on this front.
2 The third piece is acquiring the
analytics capabilities needed to derive insights from data; organizations may choose to
add in-house capabilities or outsource to specialists. The fourth component is a common
stumbling block: changing business processes to incorporate data insights into the actual
workflow. This requires getting the right data insights into the hands of the right personnel.
Finally, organizations need to build the capabilities of executives and mid-level managers to
understand how to use data-driven insights—and to begin to rely on them as the basis for
making decisions.
2 Digital America: A tale of the haves and have-mores, McKinsey Global Institute, December 2015; and Digital
Europe: Pushing the frontier, capturing the benefits
, McKinsey Global Institute, June 2016.
10-20%
of the potential
value has been
captured in the
public sector and
health care

4 McKinsey Global Institute Executive summary
Where digital native companies were built for analytics, legacy companies have to do the
hard work of overhauling or changing existing systems. Neglecting any of these elements
can limit the potential value of analytics or even leave an organization vulnerable to being
disrupted. It may be a difficult transition, but some long-established names, including GE
and Union Pacific, have managed to pull it off.
THERE IS A CONTINUING SHORTAGE OF ANALYTICS TALENT
Across the board, companies report that finding the right talent is the biggest hurdle
they face in trying to integrate data and analytics into their existing operations. In a recent
McKinsey & Company survey, approximately half of executives across geographies and
industries reported greater difficulty recruiting analytical talent than filling any other kind of
role. Forty percent say retention is also an issue.
3
Data scientists, in particular, are in high demand. Our 2011 report hypothesized that
demand for data scientists would outstrip supply. This is in fact what we see in the labor
market today, despite the fact that universities are adding data and analytics programs and
that other types of training programs are proliferating. Average wages for data scientists
in the United States rose by approximately 16 percent a year from 2012 to 2014.
4 This far
3 “The need to lead in data and analytics,” McKinsey & Company survey, McKinsey.com, April 2016, available at
http://www.mckinsey.com/business-functions/business-technology/our-insights/the-need-to-lead-in-dataand-analytics. The online survey, conducted in September 2015, garnered responses from more than 500
executives across a variety of regions, industries, and company sizes.
4 Beyond the talent shortage: How tech candidates search for jobs, Indeed.com, September 2015.
Exhibit E2
Successful data and analytics transformation requires focusing on five elements
SOURCE: McKinsey Analytics; McKinsey Global Institute analysis
USE
CASES/

SOURCES
OF VALUE
Clearly articu lating

DATA
ECOSYSTEM
MODELING
INSIGHTS
WORKFLOW
INTEGRATION ADOPTION
Internal
Data modeling
“black box”
Process redesign Capability building
External Heuristic insights

“smart box
Gatheri ng data from

Tech enablement Change management
the business need
and projected impact
Outlining a clear
vision of how the
business would use
the solution
internal systems and
external sources
Appending key
external data
Creating an analytic
“sandbox”
Enhancing data
(deriving new
predictor variables)

Applyin g linear and

nonlinear modeling
to derive new
insights
Codifying and testing
heuristics
across the
organization
(informing predictor
variables)

Redesignin g

processes
Developing an
intuitive user
interface that is
integrated into dayto-day workflow
Automating
workflows

Building frontline

and management
capabilities
Proactively
managing change
and tracking
adoption
with performance
indicators
REPEATS in report
McKinsey Global Institute The age of analytics: Competing in a data-driven world 5
outstrips the less than 2 percent increase in nominal average wages across all occupations
in US Bureau of Labor Statistics data. The scarcity of elite data scientists has even been
a factor in some acquisitions of cutting-edge artificial intelligence (AI) startups; deals can
command around $5 million to $10 million per employee.
This trend is likely to continue in the near term. While we estimate that the number of
graduates from data science programs could increase by a robust 7 percent per year, our
high-case scenario projects even greater (12 percent) annual growth in demand, which
would lead to a shortfall of some 250,000 data scientists. But a countervailing force could
ease this imbalance in the medium term: data preparation, which accounts for more than
50 percent of data science work, could be automated. Whether that dampens the demand
for data scientists or simply enables data scientists to shift their work toward analysis and
other activities remains to be seen.
Many organizations focus on the need for data scientists, assuming their presence alone
will enable an analytics transformation. But another equally vital role is that of the business
translator who serves as the link between analytical talent and practical applications to
business questions. In addition to being data savvy, business translators need to have deep
organizational knowledge and industry or functional expertise. This enables them to ask
the data science team the right questions and to derive the right insights from their findings.
It may be possible to outsource analytics activities, but business translator roles require
proprietary knowledge and should be more deeply embedded into the organization. Many
organizations are building these capabilities from within.
We estimate there could be demand for approximately two million to four million business
translators in the United States alone over the next decade. Given the roughly 9.5 million
US graduates in business and in the STEM fields of science, technology, engineering, and
mathematics expected over the same period, nearly 20 to 40 percent of these graduates
would need to go into business translator roles to meet demand.
5 Today that figure is
only about 10 percent. To reduce this mismatch, wages may have to increase, or more
companies will need to implement their own training programs.
6
As data grows more complex, distilling it and bringing it to life through visualization is
becoming critical to help make the results of data analyses digestible for decision makers.
We estimate that demand for visualization grew roughly 50 percent annually from 2010
to 2015.
7 In many instances today, organizations are seeking data scientist or business
translator candidates who can also execute visualizations. However, we expect that
medium-size and large organizations, as well as analytics service providers, will increasingly
create specialized positions for candidates who combine a strong understanding of data
with user interface, user experience, and graphic design skills.
ANALYTICS LEADERS ARE CHANGING THE NATURE OF COMPETITION AND
CONSOLIDATING BIG ADVANTAGES
There are now major disparities in performance between a small group of technology
leaders and the average company—in some cases creating winner-take-most dynamics.
Leaders such as Apple, Alphabet/Google, Amazon, Facebook, Microsoft, GE, and Alibaba
Group have established themselves as some of the most valuable companies in the world.
8
The same trend can be seen among privately held companies. The leading global “unicorns”
5 Non-STEM graduates with quantitative skills can also fill business translator roles.
6 Sam Ransbotham, David Kiron, and Pamela Kirk Prentice, “The talent dividend: Analytics talent is driving
competitive advantage at data-oriented companies,”
MIT Sloan Management Review, April 25, 2015.
7 Based on using the Burning Glass job postings database to search for postings including any of the following
skills: data visualization, Tableau, Qlikview, and Spotfire. Normalized with the total number of job postings.
8 Michael Chui and James Manyika, “Competition at the digital edge: ‘Hyperscale businesses,’” McKinsey
Quarterly
, March 2015.
2M-4M
projected US
demand for
business
translators over
the next decade

6 McKinsey Global Institute Executive summary
tend to be companies with business models predicated on data and analytics, such as
Uber, Lyft, Didi Chuxing, Palantir, Flipkart, Airbnb, DJI, Snapchat, Pinterest, BlaBlaCar, and
Spotify. These companies differentiate themselves through their data and analytics assets,
processes, and strategies.
The relative value of various assets has shifted. Where previous titans of industry poured
billions into factories and equipment, the new leaders invest heavily in digital platforms, data,
and analytical talent. New digital native players can circumvent traditional barriers to entry,
such as the need to build traditional fixed assets, which enables them to enter markets with
surprising speed. Amazon challenged the rest of the retail sector without building stores
(though it does have a highly digitized physical distribution network), “fintechs” are providing
financial services without physical bank branches, Netflix is changing the media landscape
without connecting cables to customers’ homes, and Airbnb has introduced a radical new
model in the hospitality sector without building hotels. But some digital natives are now
erecting new barriers to entry themselves; platforms may have such strong network effects
that they give operators a formidable advantage within a given market.
The leading firms have a remarkable depth of analytical talent deployed on a variety of
problems—and they are actively looking for ways to enter other industries. These companies
can take advantage of their scale and data insights to add new business lines, and those
expansions are increasingly blurring traditional sector boundaries.
9 Apple and Alibaba,
for instance, have introduced financial products and services, while Google is developing
autonomous cars. The importance of data has also upended the traditional relationship
between organizations and their customers since every interaction generates information.
Sometimes the data itself is so prized that companies offer free services in order to obtain
it; this is the case with Facebook, LinkedIn, Pinterest, Twitter, Tencent, and many others. An
underlying barter system is at work, particularly in the consumer space, as individuals gain
access to digital services in return for data about their behaviors and transactions.
THE VALUE OF DATA DEPENDS ON ITS ULTIMATE USE, AND ECOSYSTEMS
ARE EVOLVING TO HELP COMPANIES CAPTURE THAT VALUE
Data is at the heart of the disruptions occurring across the economy. It has become a critical
corporate asset, and business leaders want to know what the information they hold is worth.
But its value is tied to how it will be used and by whom. A piece of data may yield nothing,
or it may yield the key to launching a new product line or cracking a scientific question. It
might affect only a small percentage of a company’s revenue today, but it could be a driver
of growth in the future.
Not all data are created equal
Part of the challenge in valuing data is its sheer diversity. Some of the broad categories
include behavioral data (capturing actions in both digital and physical environments),
transactional data (records of business dealings), ambient or environmental data (conditions
in the physical world monitored and captured by sensors), geospatial data, reference
material or knowledge (news stories, textbooks, reference works, literature, and the like), and
public records. Some data are structured (that it, easily expressed in rows and columns),
while images, audio, and video are unstructured. Data can also come from the web, social
media, industrial sensors, payment systems, cameras, wearable devices, and human entry.
Billions of mobile phones, in particular, are capturing images, video, and location data.
On the demand side, data can provide insights for diverse uses, some of which are more
valuable than others.
9 Playing to win: The new global competition for corporate profits, McKinsey Global Institute, September 2015.
McKinsey Global Institute The age of analytics: Competing in a data-driven world 7
Over the long term, value will likely accrue to providers of analytics and data
platform owners
Many organizations are hungry to use data to grow and improve performance—and multiple
players see market opportunities in this explosion of demand. There are typically many
steps between raw data and actual usage, and there are openings to add value at various
points along the way. To simplify, we focused on three categories of players in the data
ecosystem, recognizing that some players might fill more than one role.
ƒ Data generation and collection: The source and platform where data are
initially captured.
ƒ Data aggregation: Processes and platforms for combining data from multiple sources.
ƒ Data analysis: The gleaning of insights from data that can be acted upon.
Usually, the biggest opportunities are unlikely to be in directly monetizing data. As data
become easier to collect and as storage costs go down, most data are becoming more
commoditized. Proxies now exist for data that were once scarce; Google Trends, for
instance, offers a free proxy for public sentiment data that previously would have been
collected through phone surveys.
However, there are important exceptions to the commoditization trend. When access is
limited by physical barriers or collection is expensive, data will hold its value. An important
case in which value can accrue to data generation and collection involves market-making
or social media platforms with strong network effects. In certain arenas, a small number
of players establish such critical mass that they are in a position to collect and own the
vast majority of user behavior data generated in these ecosystems. But in the absence of
these types of exceptional supply constraints, simply selling raw data is likely to generate
diminishing returns over time.
Another role in the data ecosystem involves aggregating information from different sources.
In general, this capability is becoming more accessible and less expensive, but this role can
be valuable when certain conditions apply. Data aggregation adds value when combining,
processing, and aggregating data is technically difficult or organizationally challenging
(for example, when aggregating involves coordinating access across diverse sources).
Some companies have built business models around serving as third-party aggregators
for competitors within a given industry, and this model has the potential to create network
effects as well.
The third part of the data ecosystem, analytics, is where we expect to see the biggest
opportunities in the future. The provider of analytics understands the value being generated
by those insights and is thus best positioned to capture a portion of that value. Data
analytics tools, like other software, already command large margins. Combining analytical
tools with business insights for decision makers is likely to multiply the value even further.
Increasingly complex data and analytics will require sophisticated translation, and use
cases will be very firm-specific. Bad analysis can destroy the potential value of high-quality
data, while great analysis can squeeze insights from even mediocre data. In addition, the
scarcity of analytics talent is driving up the cost of these services. Given the size of the
opportunities, firms in other parts of the ecosystem are scrambling to stake out a niche in
the analytics market. Data aggregators are offering to integrate clients’ data and perform
analysis as a service. One-stop shops offering integrated technology stacks are adding
analytics capabilities, such as IBM Watson, as are other professional services and business
intelligence firms.

8 McKinsey Global Institute Executive summary
SIX DISRUPTIVE DATA-DRIVEN MODELS AND CAPABILITIES ARE RESHAPING
SOME INDUSTRIES—AND COULD SOON TRANSFORM MANY MORE
Certain characteristics of a given market (such as inefficient matching, information
asymmetries, and human biases and errors) open the door to disruption. They set the
stage for six archetypes to have a major effect (Exhibit E3). In each of these models, the
introduction of new data is a key enabler.
Bringing in orthogonal data can change the basis of competition
As data proliferate, many new types, from new sources, can be brought to bear on any
problem. In industries where most incumbents have become used to relying on a certain
kind of standardized data to make decisions, bringing in fresh types of data sets to
supplement those already in use can change the basis of competition. New entrants with
privileged access to these “orthogonal” data sets can pose a uniquely powerful challenge
to incumbents. We see this playing out in property and casualty insurance, where new
companies have entered the marketplace with telematics data that provides insight into
driving behavior. This is orthogonal to the demographic data that had previously been used
for underwriting. Other domains could be fertile ground for bringing in orthogonal data from
the internet of things (IoT). Connected light fixtures, which sense the presence of people
in a room and have been sold with the promise of reducing energy usage, generate “data
exhaust” that property managers can use to optimize physical space planning. Even in
human resources, some organizations have secured employee buy-in to wear devices that
capture data and yield insights into the “real” social networks that exist in the workplace,
enabling these organizations to optimize collaboration through changes in work spaces.
Orthogonal data will rarely replace the data that are already in use in a domain; it is more
likely that an organization will integrate orthogonal data with existing data. Within the other
Exhibit E3
Archetype of disruption Domains that could be disrupted
Business models
enabled by orthogonal
data
ƒ Insurance
ƒ Health care
ƒ Human capital/talent
Hyperscale, real-time
matching
ƒ Transportation and logistics
ƒ Automotive
ƒ Smart cities and infrastructure
Radical personalization ƒ Health care
ƒ Retail
ƒ Media
ƒ Education
Massive data
integration capabilities
ƒ Banking
ƒ Insurance
ƒ Public sector
ƒ Human capital/talent
Data-driven discovery ƒ Life sciences and pharmaceuticals
ƒ Material sciences
ƒ Technology
Enhanced decision
making
ƒ Smart cities
ƒ Health care
ƒ Insurance
ƒ Human capital/talent
Data and analytics underpin six disruptive models, and certain characteristics make individual domains susceptible
SOURCE: McKinsey Global Institute analysis
Indicators of potential for disruption:
Assets are underutilized due to
inefficient signaling
Supply/demand mismatch
Dependence on large amounts of
personalized data
Data is siloed or fragmented
Large value in combining data from
multiple sources
R&D is core to the business model
Decision making is subject to human
biases
Speed of decision making limited by
human constraints
Large value associated with improving
accuracy of prediction
REPEATS in report
McKinsey Global Institute The age of analytics: Competing in a data-driven world 9
archetypes below are several examples of orthogonal data being combined with existing
data to create new business models and improve performance.
Hyperscale platforms can match supply and demand in real time
Digital platforms provide marketplaces that connect sellers and buyers for many products
and services. Some platform operators using data and analytics to do this in real time and
on an unprecedented scale—and this can be transformative in markets where supply and
demand matching has been inefficient.
In personal transportation, ride-sharing services use geospatial mapping technology to
collect crucial data about the precise location of passengers and available drivers in real
time. The introduction of this new type of data enabled efficient and instant matching, a
crucial innovation in this market. In addition, the data can be analyzed at the aggregate
level for dynamic pricing adjustments to help supply and demand adjust. The typical
personally owned car is estimated to sit idle 85 to 95 percent of the time, making it a hugely
underutilized asset. Platforms such as Uber, Lyft, and Chinese ride-sharing giant Didi
Chuxing have been able to expand rapidly without acquiring huge fleets themselves, making
it easy for new drivers to put their own underutilized assets to work.
By 2030 mobility services, such as ride sharing and car sharing, could account for more
than 15 to 20 percent of total passenger vehicle miles globally. This growth—and the
resulting hit to the taxi industry—may be only a hint of what is to come. Automakers are
the biggest question mark. While sales will likely continue to grow in absolute numbers, we
estimate that the shift toward mobility services could halve the growth rate of global vehicle
sales by 2030. Consumers could save on car purchases, fuel, and parking. If mobility
services attain 10 to 30 percent adoption among low-mileage urban vehicle users, the
ensuing economic impact could reach $845 billion to some $2.5 trillion globally by 2025.
Some of this value will surely go to consumer surplus, while some will go to the providers of
these platforms and mobility services.
Data and analytics enable “radical personalization”
Data and analytics can reveal finer levels of distinctions, and one of the most powerful
uses is micro-segmenting a population based on the characteristics of individuals. Using
the resulting insights to personalize products and services on a wide scale is changing the
fundamentals of competition in many sectors, including education, travel and leisure, media,
retail, and advertising.
This capability could have profound implications for the way health care is delivered if the
sector can incorporate the behavioral, genetic, and molecular data connected with many
individual patients. The declining costs of genome sequencing, the advent of proteomics,
and the growth of real-time monitoring technologies make it possible to generate this kind
of new, ultra-granular data. These data can reshape health care in two profound ways.
First, they can help address information asymmetries and incentive problems in the healthcare system. Now that a more complete view of the patient is available, incentives could
be changed for hospitals and other providers to shift their focus from disease treatment
to wellness and prevention, saving huge sums on medical expenditures and improving
the quality of life. Second, having more granular and complete data on individual patients
can make treatments more precise. Pharmaceutical and medical device companies
have enormous possibilities in R&D for accelerating drug discovery, although they will be
challenged to create new business models to deliver treatments tailored to smaller, targeted
patient populations. Treatments, dosages, and care settings can be personalized to
individuals, leading to more effective outcomes with fewer side effects and reduced costs.
Personalized medicine could reduce health-care costs while allowing people to enjoy
longer, healthier, and more productive lives. The total impact could range from $2 trillion
Up to
$2.5T
potential economic
impact from
continued adoption
of mobility services
by 2025

10 McKinsey Global Institute Executive summary
to $10 trillion. The wide range depends on the many uncertainties involved, including
how rapidly the health-care system can adapt and whether R&D applications produce
breakthrough treatments.
Massive data integration capabilities can break down organizational silos
The first step in creating value from data and analytics is accessing all the information
that is relevant to a given problem. This may involve generating the data, accessing it
from new sources, breaking silos within an organization to link existing data, or all of the
above. Combining and integrating large stores of data from all of these varied sources has
incredible potential to yield insights, but many organizations have struggled with creating the
right structure for that synthesis to take place.
Retail banking, for instance, is an industry rich with data on customers’ transactions,
financial status, and demographics. But few institutions have made the most of the data
due to internal barriers and the variable quality of the information itself. Surmounting
these barriers is critical now that social media, call center discussions, video footage from
branches, and data acquired from external sources and partners can be used to form a
more complete picture of customers. Massive data integration has significant potential for
retail banks. It can enable better cross-selling, the development of personalized products,
dynamic pricing, better risk assessment, and more effective marketing—and it can help
firms achieve more competitive cost structures than many incumbent institutions. All
told, we estimate a potential economic impact of $110 billion to $170 billion in the retail
banking industry in developed markets and approximately $60 billion to $90 billion in
emerging markets.
Additionally, companies in other sectors can become part of the financial services
ecosystem if they bring in orthogonal data—such as non-financial data that provides a more
comprehensive and detailed view of the customer. These players may have large customer
bases and advanced analytics capabilities created for their core businesses, and they can
use these advantages to make rapid moves across sector boundaries. Alibaba’s creation of
Alipay and Apple’s unveiling of Apple Pay are prime examples of this trend.
Data and analytics can fuel discovery and innovation
One of the main components of productivity growth, innovation can be applied to both
processes and products. Throughout history, innovative ideas have sprung from human
ingenuity and creativity—but now data and algorithms can support, enhance, or even
replace human ingenuity in some instances.
In the realm of process innovation, data and analytics are helping organizations determine
how to structure teams, resources, and workflows. High-performing teams can be many
times more productive than low-performing teams, so understanding this variance and how
to build more effective collaboration is a huge opportunity for organizations. This involves
looking at issues such as the complementarity of skills, optimal team sizes, whether teams
need to work together in person, what past experience or training is important, and even
how their personalities may mesh. Data and analytics can test hypotheses and find new
patterns that may not have even occurred to managers. Vast amounts of email, calendar,
locational, and other data are available to understand how people work together and
communicate, all of which can lead to new insights about improving performance.
In product innovation, data and analytics can transform research and development in areas
such as materials science, synthetic biology, and life sciences. Leading pharmaceutical
companies are using data and analytics to aid with drug discovery. Data from a variety of
sources could better determine the chemical compounds that would serves as effective
drug treatments for a variety of diseases. AstraZeneca and Human Longevity are partnering
Up to
$260B
potential global
impact of massive
data integration in
retail banking

McKinsey Global Institute The age of analytics: Competing in a data-driven world 11
to build a database of one million genomic and health records along with 500,000
DNA samples from clinical trials. The associations and patterns that can be gleaned
from that data could prove to be immensely valuable in advancing scientific and drug
development breakthroughs.
Algorithms can support and enhance human decision making
When humans make decisions, the process is often muddy, biased, or limited by our
inability to process information overload. Data and analytics can change all that by bringing
in more data points from new sources, breaking down information asymmetries, and adding
automated algorithms to make the process instantaneous. As the sources of data grow
richer and more diverse, there are many ways to use the resulting insights to make decisions
faster, more accurate, more consistent, and more transparent.
There are many examples of how this can play out in industries and domains across the
economy. Smart cities, for example, are one of the most promising settings for applying
the ability of machines and algorithms to process huge quantities of information in a
fraction of the time it takes humans. Using sensors to improve traffic flows and the internet
of things to enable utilities to reduce waste and keep infrastructure systems working at
top efficiency are just two of the myriad possible municipal applications. One of the most
promising applications of data and analytics is in the prevention of medical errors. Advanced
analytical support tools can flag potential allergies or dangerous drug interactions for
doctors and pharmacists alike, ensuring that their decisions are consistent and reliable.
And finally, perhaps no area of human decision making is quite as opaque and clouded by
asymmetric information as hiring. Data and analytics have the potential to create a more
transparent labor market by giving employers and job seekers access to data on the supply
and demand for particular skills, the wages associated with various jobs, and the value of
different degree programs.
THE FRONTIERS OF MACHINE LEARNING, INCLUDING DEEP LEARNING,
HAVE RELEVANCE IN EVERY INDUSTRY AND WIDE-RANGING POTENTIAL TO
SOLVE PROBLEMS
Machine learning can enhance the power of each of the archetypes described above.
Conventional software programs are hard-coded by humans with specific instructions on
the tasks they need to execute. By contrast, it is possible to create algorithms that “learn”
from data without being explicitly programmed. The concept underpinning machine learning
is to give the algorithm a massive number of “experiences” (training data) and a generalized
strategy for learning, then let it identify patterns, associations, and insights from the data. In
short, these systems are trained rather than programmed.
Some machine learning techniques, such as regressions, support vector machines, and
k-means clustering, have been in use for decades. Others, while developed previously,
have become viable only now that vast quantities of data and unprecedented processing
power are available. Deep learning, a frontier area of research within machine learning,
uses neural networks with many layers (hence the label “deep”) to push the boundaries of
machine capabilities. Data scientists have recently made breakthroughs using deep learning
to recognize objects and faces and to understand and generate language. Reinforcement
learning is used to identify the best actions to take now in order to reach some future goal.
These type of problems are common in games but can be useful for solving dynamic
optimization and control theory problems—exactly the type of issues that come up
in modeling complex systems in fields such as engineering and economics. Transfer
learning focuses on storing knowledge gained while solving one problem and applying it
to a different problem. Machine learning, combined with other techniques, could have an
enormous range of uses (see Exhibit E4 and Box E1, “The impact of machine learning”).

12 McKinsey Global Institute Executive summary
This research offers a broad initial exploration of machine learning through two lenses. First,
we investigate which business uses across 12 industries could be met by machine learning.
Second, we examine which work activities currently performed by people could potentially
be automated through machine learning and how that could play out across occupations.
The initial findings here are meant to set the stage for future research.
Understanding the capabilities of machine learning and deep learning
Machine learning capabilities are best suited for solving three broad categories of problems:
classification, prediction/estimation, and generation (Exhibit E5). Classification problems
are about observing the world, including identifying objects in images and video, and
recognizing text and audio. Classification also involves finding associations in data or
segmenting it into clusters, which is useful in tasks such as customer segmentation.
Machine learning can also be used to predict the likelihood of events and forecast
outcomes. Lastly, it can be used to generate content, from interpolating missing data to
generating the next frame in a video sequence.
Exhibit E4
SOURCE: McKinsey Global Institute analysis
Machine learning can be combined with other types of analytics to solve a large swath of business problems
Machine learning techniques
(not exhaustive)
Clustering
(e.g., k-means)
Dimensionality reduction
Classification
(e.g., support vector machines)
Conventional neural networks
Deep learning networks
Convolutional neural network
Recurrent neural network
Deep belief networks
Problem types
Classification
Prediction
Generation
Other analytics
(not exhaustive)
Search algorithms
Sorting
Merging
Compression
Graph algorithms
Linear and non-linear
optimization
Signal processing
Encryption
Use cases
Resource allocation
Predictive analytics
Predictive maintenance
Hyper-personalization
Discover new trends/anomalies
Forecasting
Price and product optimization
Convert unstructured data
Triaging
REPEATS in report
Regression
(e.g., logistic)

McKinsey Global Institute The age of analytics: Competing in a data-driven world 13
Exhibit E5
SOURCE: McKinsey Global Institute analysis
Classification Classify/label visual objects Identify objects, faces in images and video
Classify/label writing and text Identify letters, symbols, words in writing sample
Classify/label audio Classify and label songs from audio samples
Cluster, group other data Segment objects (e.g., customers, product features) into
categories, clusters
Discover associations Identify that people who watch certain TV shows also read certain
books
Prediction Predict probability of outcomes Predict the probability that a customer will choose another provider
Forecast Trained on historical data, forecast demand for a product
Value function estimation Trained on thousands of games played, predict/estimate rewards
from actions from future states for dynamic games
Generation Generate visual objects Trained on a set of artist’s paintings, generate a new painting in the
same style
Generate writing and text Trained on a historical text, fill in missing parts of a single page
Generate audio Generate a new potential recording in the same style/genre
Generate other data Trained on certain countries’ weather data, fill in missing data
points for countries with low data quality
Machine learning can help solve classification, prediction, and generation problems
REPEATS in report
Box E1. The impact of machine learning
Machine learning can be applied to a tremendous variety
of problems—from keeping race cars running at peak
performance to ferreting out fraud.
Off the track, Formula One (F1) teams compete in an
arms race to make their cars faster. Top F1 teams pour
hundreds of millions of dollars annually into development,
continually aiming for incremental technological
improvements that can boost speed. With so much at
stake, F1 engineering teams constantly seek to improve
productivity. Three F1 teams recently turned to machine
learning to hold down costs in their aerodynamics
operations divisions, which typically eat up more than
80 percent of development resources. Building on years
of diverse project data—including CAD logs, human
resources data, and employee communications—they
looked for patterns that influenced the efficiency of an
individual project. They discovered, for example, that too
many engineers or long stoppages typically increased
labor hours on a given project by 5 to 6 percent, while
team use of the documentation system improved
productivity by more than 4 percent. Overall, this
application reduced the budget by 12 to 18 percent,
saving millions of dollars.
Another application of machine learning, predictive
analytics, has proven to be effective at spotting fraud. At
one large auto insurer, high accident rates for new client
policies suggested that claims were being filed for preexisting damage. The machine learning model was able
to use diverse data to identify groups of new policies
with accident rates six times those of the median. This
grouping formed the basis of a new pricing strategy
that improved profitability by more than 10 percent.
Separately, a large retail bank in the United Kingdom
used machine learning algorithms to identify fraudulent
transactions with more than 90 percent accuracy. In
another example, a large payment processor deployed
machine learning on its extensive transaction data to
identify “mule accounts” involved in money laundering.

14 McKinsey Global Institute Executive summary
We identified 120 potential use cases of machine learning in 12 industries, and surveyed
more than 600 industry experts on their potential impact. The most striking finding was
the extraordinary breadth of the potential applications of machine learning; each of the use
cases was identified as being one of the top three in an industry by at least one expert in that
industry. But there were differences.
We plotted the top 120 use cases in Exhibit E6. The y-axis shows the volume of available
data (encompassing its breadth and frequency), while the x-axis shows the potential impact,
based on surveys of more than 600 industry experts. The size of the bubble reflects the
diversity of the available data sources.
Exhibit E6
SOURCE: McKinsey Global Institute analysis
Machine learning has broad potential across industries and use cases
0 0.1 0.9 1.2 1.3 1.5 1.9
9.5
1.1 1.7
3.5
3.0
2.5
1.6
2.0
9.0
10.0
8.5
8.0
7.0
6.0
6.5
7.5
5.5
5.0
4.5
4.0
1.0
1.5
0.5
0.2 0.4
1.0
0.3 0.5 0.6 0.7 0.8 1.4 1.8
0

Predictive maintenance
Predictive
maintenance
(energy)
Optimize pricing
and scheduling
in real time
Personalize crops to
individual conditions
Identify
fraudulent
transactions
Discover new
consumer trends
Predict
personalized
health outcomes
Identify and
navigate roads
Personalize
financial
products
Personalize
advertising
Optimize
merchandising
strategy
Higher potential
Lower priority
(manufacturing)
Diagnose diseases
Optimize clinical trials

Volume
Breadth and frequency of data
Impact score
Consumer
Agriculture
Health care
Finance
Energy
Automotive
Manufacturing Public/social
Pharmaceuticals
Media
Travel, transport,
and logistics
Telecom
Size of bubble indicates
variety of data (number
of data types)
S in report
McKinsey Global Institute The age of analytics: Competing in a data-driven world 15
The industry-specific uses that combine data richness with a larger opportunity are
the largest bubbles in the top right quadrant of the chart. These represent areas where
organizations should prioritize the use of machine learning and prepare for a transformation
to take place. Some of the highest-opportunity use cases include personalized advertising;
autonomous vehicles; optimizing pricing, routing, and scheduling based on real-time data in
travel and logistics; predicting personalized health outcomes; and optimizing merchandising
strategy in retail.
The use cases in the top right quadrant fall into four main categories. First is the radical
personalization of products and services for customers in sectors such as consumer
packaged goods, finance and insurance, health care, and media—an opportunity that
most companies have yet to fully exploit. The second is predictive analytics. This includes
examples such as triaging customer service calls; segmenting customers based on
risk, churn, and purchasing patterns; identifying fraud and anomalies in banking and
cybersecurity; and diagnosing diseases from scans, biopsies, and other data. The third
category is strategic optimization, which includes uses such as merchandising and shelf
optimization in retail, scheduling and assigning frontline workers, and optimizing teams
and other resources across geographies and accounts. The fourth category is optimizing
operations and logistics in real time, which includes automating plants and machinery to
reduce errors and improve efficiency, and optimizing supply chain management.
Advances in deep learning could greatly expand the scope of automation
Previous MGI research examined the potential to automate 2,000 work activities performed
in every occupation in the economy.
10 For each work activity, we identified the required
level of machine performance across 18 human capabilities that could potentially
enable automation.
Machine learning is particularly well-suited to implement seven of those 18 capabilities
(Exhibit E7). The first striking observation is that almost all activities require some capabilities
that correlate with what machine learning can do. In fact, only four out of more than 2,000
detailed work activities (or 0.2 percent) do not require any of the seven machine learning
capabilities. Recognizing known patterns, by itself, is needed in 99 percent of all activities
to varying degrees. This is not to say that such a high share of jobs is likely to be automated,
but it does underscore the wide applicability of machine learning in many workplaces.
MGI’s previous research on automation found that 45 percent of all work activities,
associated with $14.6 trillion of wages globally, have the potential to be automated
by adapting currently demonstrated technology. Some 80 percent of that could be
implemented by using existing machine learning capabilities. But deep learning is in its early
stages. Improvements in its capabilities, particularly in natural language understanding,
suggest the potential for an even greater degree of automation. In 16 percent of work
activities that require the use of language, for example, increasing the performance of
machine learning in natural language understanding is the only barrier to automation.
Improving natural language capabilities alone could lead to an additional $3 trillion in
potential global wage impact.
10 These “detailed work activities” are defined by O*NET, a data collection program sponsored by the US
Department of Labor. See Michael Chui, James Manyika, and Mehdi Miremadi, “Four fundamentals of
workplace automation,”
McKinsey Quarterly, November 2015.
$3T
wages potentially
affected if machine
learning gains
better capabilities
in natural language
understanding

16 McKinsey Global Institute Executive summary
We further looked at which occupations that could be affected by improvements in deep
learning represent the greatest potential wage impact (Exhibit E8). The role of customer
service representatives, in particular, lends itself to automation across most of its work
activities. Deep learning is also likely to have a large impact on frontline supervisory roles
and in occupations with primarily administrative duties, including executive assistants,
cashiers, and waitstaff. Large numbers of people are employed in these occupations, which
points to the possibility of substantial job displacement. In addition, advances in machine
learning could automate significant percentages of the activities associated with some highpaying jobs such as lawyers and nurses.
While machine learning in general and deep learning in particular have exciting and wideranging potential, there are real concerns associated with their development and potential
deployment. Some of these, such as privacy, data security, and data ownership, were
present even before the big data age. But today new questions have formed.
Exhibit E7
SOURCE: McKinsey Global Institute analysis
Improvements in natural learning understanding and generation as well as social sensing would have the biggest
impact on expanding the number of work activities that deep learning could technically automate

79
33
25
20
59
76

99
Generating novel
patterns/categories
Natural language
generation
Optimization
and planning
Social and
emotional sensing
Recognizing known
patterns/categories
Sensory perception
Natural language
understanding
4

16
41
5
25
1
2

53
Share of detailed work activities
(DWAs) that require this capability
Where required, share of DWAs where
current level is inadequate
Capability
%
% of DWAs where this capability
is the only gap
REPEATS in report
McKinsey Global Institute The age of analytics: Competing in a data-driven world 17
Exhibit E8

31
Providing consultation and
advice to others 2.3 41.14 61.8
Lawyers
51
Performing for or working
directly with the public 6.9 9.35 67.4
35
Interpreting the meaning of
information for others 12.8 8.75 81.5
First-line supervisors of
office and administrative
support workers
30
Performing administrative
activities 6.1 24.68 94.2
Business operations
specialists, all other
13
Guiding, directing, and
motivating subordinates 19.7 15.02 77.4
First-line supervisors of
retail sales workers
28
Interacting with computers
to enter data, process
information, etc.
48.2 3.90 109.8
Secretaries and
administrative
assistants, except legal,
medical, and executive
20
Industrial engineers Getting information 8.0 20.60 69.4
24
Organizing, planning, and
prioritizing work 8.5 12.73 54.2
First-line supervisors of
helpers, laborers, and
material movers
27
Monitoring processes,
materials, or surroundings 8.3 18.25 86.7
Managers, all other
18
Performing administrative
activities 68.1 3.18 81.5
Cashiers
Customer service
representatives

Occupations
% of time spent on activities
that could be automated if DL
improves (by DWA group)
1
Most frequently
performed group of
DWAs that could be
automated if DL improves
Global
employment
Million
Hourly
wage
$
Global
wages that
DL could
automate
$ billion
SOURCE: National labor and statistical sources; McKinsey Global Institute analysis
Improvements in deep learning (DL) could affect billions of dollars in wages in ten occupations globally
1 Detailed work activity. There are 37 total DWA groups.
Social and
emotional sensing
Sensory
perception
Natural language
understanding
Generating novel
patterns/categories
Multiple

18 McKinsey Global Institute Executive summary
First, deep learning models are opaque, which can be a barrier to adoption in certain
applications. As of today, it is difficult to decipher how deep neural networks reach insights
and conclusions, making their use challenging in cases where transparency of decision
making may be needed for regulatory purposes. Also, decision makers and customers may
not buy into insights generated in a non-transparent way, especially when those insights
are counterintuitive.
Second, there are ethical questions surrounding machine intelligence. One set of ethical
concerns relates to real-world biases that might be embedded into training data. Another
question involves deciding whose ethical guidelines will be encoded in the decision making
of intelligence and who is responsible for the algorithm’s conclusions. Leading artificial
intelligence experts, through OpenAI, the Foundation for Responsible Robotics, and other
efforts, have begun tackling these questions.
Third, the potential risks of labor disruption from the use of deep learning to automate
activities are generating anxiety. There is historical precedent for major shifts among sectors
and changes in the nature of jobs in previous waves of automation. In the United States,
the share of farm employment fell from 40 percent in 1900 to 2 percent in 2000; similarly,
the share of manufacturing employment fell from roughly 25 percent in 1950 to less than
10 percent in 2010. In both circumstances, while some jobs disappeared, new ones were
created, although what those new jobs would be could not be ascertained at the time. But
history does not necessarily provide assurance that sufficient numbers of new, quality jobs
will be created at the right pace. At the same time, many countries have or will soon have
labor forces that are declining in size, requiring an acceleration of productivity to maintain
anticipated rates of economic growth. But automation technologies will not be widely
adopted overnight; in fact, a forthcoming MGI research report will explore the potential
pace of automation of different activities in different economies. Certainly dealing with job
displacement, retraining, and unemployment will require a complex interplay of government,
private sector, and educational and training institutions, and it will be a significant debate
and an ongoing challenge across society.
•••
Data and analytics have even greater potential to create value today than they did when
companies first began using them. Organizations that are able to harness these capabilities
effectively will be able to create significant value and differentiate themselves, while others
will find themselves increasingly at a disadvantage.

McKinsey Global Institute The age of analytics: Competing in a data-driven world 19
© Cultura Creative/Alamy
20 McKinsey Global Institute Executive summary
© Real444/Getty Images