Report from the McKinsey Technical Institute

The speed with which automation technologies are emerging, and the extent to which they could disrupt the world of work, may appear daunting but is not unprecedented. Technological change has reshaped the workplace continually over the past two centuries since the Industrial Revolution, and even earlier (see Box 1, “What history teaches us about the effect of technological change on work, employment, and productivity”). Nonetheless, the latest technological developments will touch every job in every sector and in every country. The advances we described in the previous chapter all have practical applications in the workplace, and in some cases they have already been adapted, integrated, and deployed. Sophisticated machines are replacing human labor in workplaces from factories to fast-food restaurants. They are becoming a part of everyday life in fields from journalism to law to medicine; at the University of Tokyo, for example, IBM’s Watson made headlines in 2016 by diagnosing in a 60-year-old woman a rare form of leukemia that had eluded her doctors for months. 40

These technologies bring with them progress, productivity improvements, increased efficiencies, safety, and convenience, but they also raise difficult questions about the broader impact of automation on the workforce as a whole. Think tanks and organizations such as the World Economic Forum are forecasting the likelihood of major job substitution by automation. 41 Some academic studies estimate that close to 50 percent of US and European jobs could be automated, although other studies put that figure much lower (see the table “Overview of select recent studies on the impact of automation and the future of
work,” page 21).

Some of these projections focus on occupations perceived to be at risk.42 Our approach is substantially different: we consider work activities a more relevant and useful basis for analysis than occupations. The reason for this is that, within sectors, every occupation consists of a number of constituent activities that may have a different technical potential for automation. A typical retail salesperson, for example, will spend some time interacting with customers, stocking shelves, or ringing up sales. Machines can already outperform humans in some of these activities—they are highly adept at managing warehouse inventory, for example—and at least one fashion company has a bot that advises clients, via their mobile phones, about the best lipstick match. 43 But computers and machines are far less adept than humans at sensing the emotional state of customers or understanding context. For example, no robot yet has the capacity to sense a distressed client and propose offering him or her a glass of water or a cup of tea.

40. Bernie Monegain, “IBM Watson pinpoints rare form of leukemia after doctors misdiagnosed patient,” Healthcare IT News, August 8, 2016.
41. The World Economic Forum has predicted that more than five million jobs could be lost to robots in 15 major developed and emerging economies over the next five years. The future of jobs: Employment, skills, and workforce strategy for the fourth Industrial Revolution, World Economic Forum, January 2016.
42. The Oxford University study, for example, focuses on occupations that it categorizes as being susceptible to automation. Carl Benedikt Frey and Michael A. Osborne, The future of employment: How susceptible are jobs to computerization? Oxford Martin School, September 17, 2013.
43. “Sephora debuts two new bot-powered beauty tools for Messenger,” PR Newswire, November 2, 2016.

Box 1. What history teaches us about the effect of technological change on work, employment, and productivity

The fear of technological innovation destroying jobs and displacing workers dates back several hundred years, even before the Luddite movement in Britain during the Industrial Revolution that gave its name to militant technophobia. The Luddites were textile mill workers in Nottingham who rioted in 1811 to destroy the new automated looms that threatened their livelihoods. Ever since, there has been no shortage of predictions that machines would replace human laborers, with possibly dire effects. Karl Marx wrote in 1858 that “the means of labor passes through different metamorphoses, whose culmination is the machine, or rather, an automatic system of machinery.” 1

In 1930, the British economist John Maynard Keynes coined the term “technological unemployment” to describe a situation in which innovation that economized the use of labor outstripped the pace at which new jobs could be created. Keynes warned that this was akin to a “new disease”—but he also described this malady as being a “temporary phase of maladjustment.” 2

More recently, in 1966, a report from the US National Commission on Technology, Automation, and Economic Progress, predicted that “in the new technology, machines and automated processes will do the routine and mechanical work. Human resources will be released and available for new activities beyond those that are required for mere subsistence. The great need is to discover the nature of this new kind of work, to plan it, and to do it. In the longer run, significant changes may be needed in our society—in education, for example—to help people find constructive and rewarding ways to use increasing leisure.” 3

One lesson of history is that deployment of new technologies in the past has led to new forms of work, including in cases when shifts in the activities performed in the workplace have been very substantial. In the United States, for example, the share of farm employment fell from 40 percent in 1900 to 2 percent in 2000; similarly, the share of manufacturing employment fell from 25 percent in 1950 to less than 10 percent in 2010. 4 In both cases, while some jobs disappeared, new ones were created, although what those new jobs would be could not be predicted at the time.

Technological innovation can create new demand and whole new industries. Printing is one example. When the Times of London in 1814 switched to a revolutionary steam-powered printing press invented by German engineer Friedrich Koenig, the newspaper’s printers staged a revolt that was quelled only when the paper promised to keep on displaced workers. That prototype, which used steam from water heated by coal to drive the press, initially printed 1,100 pages per hour, or five times as many as the mechanical press that preceded it. By 1820, presses could print 2,000 sheets per hour. By 1828, that doubled to 4,000. Then came the invention of rotary presses, which in turn enabled huge rolls of paper to be loaded into the presses rather than individual sheets. By the 1860s, the most advanced presses could print 30,000 pages per hour. The arrival of electricity and the development of linotype and photomechanical processes able to reproduce photographs meant that by 1890, the New York Herald was able to print 90,000 copies of its four-page paper per hour, with color illustrations. This stream of innovation, combined with greater press freedom, drove the growth of a vibrant and fast-growing newspaper industry in the United States and Europe, creating millions of jobs in printing, journalism, and other related fields. 5

More recent evidence at a macroeconomic level suggests the positive links between technological progress, productivity, and jobs continued through the 20th century. Positive gains in both productivity and employment have been reported in the United States in more than two-thirds of the years since 1929. 6 One-third of new jobs created in the United States in the past 25 years did not exist, or barely existed, 25 years ago. 7 However, in recent years, there has been a notable divergence between productivity and pay, and the labor share of income has declined in many advanced economies. 8

The question today is whether this latest wave of innovation is by its nature substantially different from technological disruptions in the past. As automation makes inroads into the workplace, a critical concern is that technology-enabled automation could replace not just low-skill jobs—which is what happened in the past—but that it could affect all jobs. For now, there is an increasing bifurcation in the labor market between a dwindling number of high-skill jobs and many low-wage and low-skill service jobs. 9 As we detail later in this chapter, even high-paying occupations in sectors such as financial services are potentially susceptible to automation. Opinions are sharply divided about the medium- and long-term effects of this automation wave. In 2014, the Pew Research Center conducted a survey of technology professionals and economists and found that 48 percent of respondents believed new technologies would displace more jobs than they would create by 2025. 10

1. Karl Marx, Grundrisse: Foundations of the critique of political economy, 1858 (unpublished manuscript), available online at
2. John Maynard Keynes, “Economic possibilities for our grandchildren,” in Essays in Persuasion, Macmillan, 1933. The essay is available online at
3. Technology and the American economy: Report of the National Commission on Technology, Automation, and Economic Progress, US Department of Health, Education, and Welfare, 1966.
4. Stanley Lebergott, “Labor force and employment 1800–1960,” in Output, employment, and productivity in the United States after 1800, Dorothy S. Brady, ed., NBER, 1966; World Bank data.
5. Elizabeth Eisenstein, The printing press as an agent of change, Cambridge University Press, 1980; Robert Hoe, A short history of the printing press and of the improvements in printing machinery from the time of Gutenberg up to the present day, 1902.
6. Growth and renewal in the United States: Retooling America’s economic engine, McKinsey Global Institute, February 2011.
7. Ibid.
8. Josh Bivens and Lawrence Mishal, Understanding the historic divergence between productivity and a typical worker’s pay: Why it matters and why it’s real, Economic Policy Institute briefing paper number 406, September 2015; Poorer than their parents? Flat or falling incomes in advanced economies, McKinsey Global Institute, July 2016.
9. See David H. Autor and David Dorn, “The growth of low-skill service jobs and the polarization of the US labor market,” American Economic Review, volume 103, number 5, August 2013, and Lawrence Mishel and Kar-Fai Gee, “Why aren’t workers benefiting from labour productivity growth in the United States?” International Productivity Monitor, number 23, spring 2012.
10. Aaron Smith and Janna Anderson, AI, robotics, and the future of jobs, Pew Research Center, August 6, 2014. There has been a proliferation of books by competing schools of “technooptimists” and “techno-pessimists.” They notably include Erik Brynjolfsson and Andrew McAfee, The second machine age: Work, progress, and prosperity in a time of brilliant technologies, W. W. Norton & Company, 2014; and Robert Gordon, The rise and fall of American growth: The US standard of living since the Civil War, Princeton University Press, 2016; Martin Ford, Rise of the robots: Technology and the threat of a jobless future, Basic Books, 2015. Also see Jason Furman, “Is this time different? The opportunities and challenges of artificial intelligence,” remarks at AI Now: The Social and Economic Implications of Artificial Intelligence Technologies in the Near Term conference in New York, July 7, 2016.

Using data from the US Bureau of Labor Statistics and O*Net, we have examined in detail more than 2,000 work activities for more than 800 occupations across the entire economy. We estimated the amount of time spent on these activities and the technical feasibility of automating each of them by adapting currently demonstrated technology. Having undertaken this analysis of the US economy, we extended our study to 45 other countries, using the most comparable data available in each. 44 This detailed research enables us to draw important conclusions about the technical feasibility of automation for the global economy today, as well as for individual professions within specific sectors, from US mortgage brokers to Indian farmers.

Our core findings are that the proportion of occupations that can be fully automated by adapting currently demonstrated technology—in other words, all of their activities could be automated—is very small, less than 5 percent in the United States. Automation will nonetheless affect almost all occupations, not just factory workers and clerks, but also landscape gardeners and dental lab technicians, fashion designers, insurance sales representatives, and CEOs, to a greater or lesser degree. The automation potential of these occupations depends on the types of work activity that they entail, but as a rule of thumb, about 60 percent of all occupations have at least 30 percent of activities that are technically automatable.

In the United States, the country for which we have the most complete data, about 46 percent of time spent on work activities across occupations and industries is technically automatable based on currently demonstrated technologies. Exhibit 2 shows the distribution range of this automation potential in the United States. On a global scale, we calculate that the adaptation of currently demonstrated automation technologies could affect 49 percent of working hours in the global economy. This potential corresponds to the equivalent of 1.1 billion workers and $11.9 trillion in wages. Among countries, the potential ranges between 40 and 55 percent, with just four countries—China, India, Japan, and the United States—accounting for just over half the total wages and workers. The potential could also be large in Europe: according to our analysis, the equivalent of 54 million full-time workers and more than $1.9 trillion in wages are associated with technically automatable activities in the continent’s five largest economies alone—France, Germany, Italy, Spain, and the United Kingdom.

In this chapter we describe in detail the technical potential for automation in different sectors of the economy and for the global economy as a whole, based on the state of technology today. We explain how we calculate technical automation potential and, based on that methodology, we identify categories of activities that are the most and the least susceptible to automation. This enables us to provide detailed estimates of the technical automation potential of sectors and of different occupations within those sectors. We conclude with an examination of similarities and differences between countries, both advanced and emerging economies, for a global view of automation and its very substantial potential to transform the world of work.

44. For full details of our methodology, see the technical appendix.

Exhibit 2

While few occupations are fully automatable, 60 percent of all occupations have at least 30 percent technically automatable activities.

Automation potential based on demonstrated technology of occupation titles in the United States (cumulative) 1

Exhibit 2

1. We define automation potential according to the work activities that can be automated by adapting currently demonstrated technology.

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis


Humans at work carry out a wide variety of activities—whether stapling documents, examining spreadsheets, meeting clients, interviewing potential new recruits, lifting crates in a store, or planting corn in a field—without consciously analyzing the exact skill sets they are using. In fact, each of these actions requires a combination of innate or acquired capabilities, ranging from manual dexterity to social perceptiveness. In order to understand and map performance requirements for machines in the workplace, we developed a detailed framework of 18 human capabilities, including physical, emotional, and cognitive ones.

It is important to note that when we discuss automation potential in this chapter, we refer to the technical potential for automation by adapting technologies that have already been demonstrated. As the technology becomes more advanced, that potential will also evolve. While this chapter focuses on technologies that have been developed today, later chapters estimate the speed with which the technological capabilities are likely to improve and be adopted in the workplace.

It should also be noted that technical feasibility is not the same as actual implementation of automation, nor a complete predictor that an activity will be automated. A second factor to consider is the cost of developing and deploying both the hardware and the software solutions for automation. The cost of labor and related supply and demand dynamics represent a third factor: if workers are in abundant supply and significantly less expensive than automation, this could be a decisive argument against adoption. A fourth factor to consider is the benefits beyond labor substitution, including higher levels of output, better quality, and fewer errors. The economic value of these benefits is often larger than that of reducing labor costs. Regulatory and social acceptance issues, such as the degree to which machines are acceptable in any particular setting, must also be weighed. A robot may, in theory, be able to replace some of the functions of a nurse, for example. But for now, the prospect that this might actually happen in a highly visible way could prove unpalatable for many patients, who expect and trust human contact. The pace at which automation will take hold in a sector or occupation reflects a subtle interplay between these factors and the trade-offs among them.

A framework of capabilities to understand the performance requirements of work activities

The framework of 18 capabilities that we developed to assess automation potential addresses a wide range of performance requirements. Many of these capabilities correspond to the technologies we discussed in the previous chapter. They cover five areas: sensory perception, cognitive capabilities, natural language processing, social and emotional capabilities, and physical capabilities (Exhibit 3).

  • Sensory perception. This includes visual perception, tactile sensing, and auditory sensing, and involves complex external perception through integrating and analyzing data from various sensors in the physical world.
  • Cognitive capabilities. A range of capabilities is included in this category including recognizing known patterns and categories (other than through sensory perception); creating and recognizing novel patterns and categories; logical reasoning and problem solving using contextual information and increasingly complex input variables; optimization and planning to achieve specific objectives given various constraints; creating diverse and novel ideas or a novel combination of ideas; information retrieval, which involves searching and retrieving information from a large range of sources; coordination with multiple agents, which involves interacting with other machines and with humans to coordinate group activity; and output articulation and presentation, which involves delivering outputs other than through natural language. These could be automated production of pictures, diagrams, graphs, or mixed media presentations.
  • Natural language processing. This consists of two distinct parts: natural language generation, which is the ability to deliver spoken messages, including with nuanced human interaction and gestures, and natural language understanding, which is the comprehension of language and nuanced linguistic communication in all its rich complexity.
  • Social and emotional capabilities. This consists of three types of capability: social and emotional sensing, which involves identifying a person’s social and emotional state; social and emotional reasoning, which entails accurately drawing conclusions based on a person’s social and emotional state, and determining an appropriate response; and social and emotional output, which is the production of an appropriate social or emotional response, both in words and through body language.
  • Physical capabilities. This includes gross motor skills, fine motor skills, navigation, and mobility. These capabilities could be implemented by robots or other machines manipulating objects with dexterity and sensitivity, moving objects with multidimensional motor skills, autonomously navigating in various environments, and moving within and across various environments and terrain.

Exhibit 3

Current technologies have achieved different levels of human performance across 18 capabilities.

Exhibit 3

1. Assumes technical capabilities demonstrated in commercial products, R&D, and academic settings; compared against human performance.

SOURCE: McKinsey Global Institute analysis

We estimated the level of performance in each of these capabilities that is required to successfully perform each work activity, categorizing whether the capability is required at all, and if so, whether the required level of performance is at roughly a median human level, below median human level, or at a high level of performance (for example, the top 25th percentile). 45 We also assessed the performance of existing technologies today against the same criteria.

This framework enabled us to assess the state of technology today and the potential to automate work activities in all sectors of the economy by adapting currently demonstrated technologies. By evaluating the technologies across a spectrum of performance, we have also been able to take into account their potential evolution in the future and resulting incremental effect on workplace activities. A detailed account of our methodology is contained in the technical appendix. Our assessments are simplifications for modeling purposes that synthesize a variety of subcapabilities, not all of which consistently fall into the below-median, median, and top-quartile categories.

Machines will need to be able to use many of these capabilities together in the workplace, as humans do

The 18 capabilities we have identified should not be taken in isolation. They are closely interconnected.

Let us return to our retail salesperson, by way of example, to show the interplay of these capabilities. Daily activities may include greeting customers, answering questions about products and services, cleaning and maintaining work areas, demonstrating product features, and processing sales and transactions. To carry out this range of activities requires almost the full spectrum of these capabilities. It starts with the greeting of customers. A skilled salesperson will identify the social and emotional state of a customer, accurately draw conclusions about how to react to that social and emotional state, and through body language, tone of voice, and choice of vocabulary, provide an emotionally appropriate response. Cognitive capabilities will be fully used, too. Listening to what a customer says and responding requires the ability to understand and generate natural language. Other cognitive capabilities employed are the ability to retrieve information (“do we have these shoes in stock?”); to reason logically and solve problems (“we don’t have them in your size in black, but we do have them in red or brown”); to coordinate with multiple agents (“I’ll have one of my colleagues determine if we have the item in stock”), and creativity (“try the purple pair, they’re very fashionable this year and will suit you well”). Physical capabilities are likewise also needed. They include mobility and navigation (walking to the stockroom), gross motor skills (taking the shoe box off a shelf in the stockroom), and fine motor skills (tying a lace).

For now, automation technologies do not have this full range of capabilities to perform at the same level as humans, and for many problems, they have yet to be seamlessly integrated into solutions. Combinations of social, cognitive, and physical capabilities are required for many activities, and we do see patterns about which capabilities are often required together (Exhibit 4). For example, an activity that needs any social or emotional capability typically needs all of them—sensing, reasoning, and output. Likewise, an activity requiring one form of physical capability such as fine motor skills or mobility tends to require multiple physical capabilities, including gross motor skills and navigation. However, cognitive capabilities can occur in combination with many different capabilities. There is an overlap between creativity and optimization and planning, for example, but one can do without the other. Coordinating with multiple agents does not automatically entail information retrieval, logical reasoning, or generating novel patterns. Moreover, where there are cognitive demands, physical demands are less likely to be required.

45. Among the many factors that define levels of performance are the acceptable error rates, particularly for sensory and cognitive activities. For instance, the consequences of false positives or negatives when making certain law enforcement or health-care judgments could be much more significant than for other judgments in the entertainment industry.

Exhibit 4

Many capabilities tend to be required together for specific activities.

Exhibit 4

SOURCE: McKinsey Global Institute analysis

When machines can take on workplace activities, the nature of work will change. Today, only about 10 percent or less of the average human worker’s time at work is spent using capabilities such as emotional reasoning and creativity, which many people would describe as being a core part of the human experience. The capability most used is recognizing known patterns, followed by natural language generation (for example, speaking), sensory perception, information retrieval, and natural language understanding (Exhibit 5). By allowing machines to handle more mundane activities, automation could free up women and men to use their creative and other talents more than they do now.

Exhibit 5

Recognizing known patterns and natural language generation are the two most-used capabilities in work activities.

Exhibit 5

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

Gauging the automation potential of occupations and sectors of the economy through analysis of their constituent activities

Applying our methodology to the more than 2,000 activities across all sectors of the economy carried out by the US workforce—and subsequently adopting similar methodology for the global workforce—we found that many types of activities share common characteristics that are readily grouped into categories. For example, processing data is a very frequent activity common to a large range of occupations in different sectors, as is carrying out repetitive physical movement. By analyzing the amount of time spent on each of these categories of activity, we were able to estimate the technical automation potential of hundreds of occupations across the economy. We have also analyzed the implications across wage rates. As noted, the technical ability to automate is only one element that will lead to automation actually being deployed in the workplace; given hardware and software costs and relative wage levels, a coherent business case needs to be made, and regulatory, social, and organizational issues also play a role.

Hourly wage rates are not strong predictors of automation potential Occupations across the spectrum of the economy, from CEO to metal welders, have a range of automation potential based on today’s technologies. Technically automatable activities represent about $2.7 trillion of addressable wages in the United States, or about 46 percent of the total hours worked. Automation is sometimes depicted as primarily affecting particular groups of workers depending on their wage levels. Our analysis finds that while there is a negative correlation between wage rates and technical automation potential, there is a large amount of variation, so the hourly wage rate is not a strong predictor of technical automation potential.46 In fact, a significant proportion of highly paid work, not just low-wage work, can be automated (Exhibit 6).

Exhibit 6

Both low and high-wage occupations have significant technical automation potential.

Exhibit 6

1. Our analysis used “detailed work activities,” as defined by O*NET, a program sponsored by the US Department of Labor, Employment and Training Administration.

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

People in the lowest wage group, earning less than $15 per hour, carry out work activities that have among the highest potential for automation, and as a whole a majority (51 percent) of this group’s activities is automatable. However, those earning between $15 and $30 per hour have an automation potential of 46 percent, which is close to the average for the economy as a whole. Above this $15 to $30 wage level, there is no clear pattern or correlation between wages and the automation potential of the work. More than 17 million American workers earn between $30 and $45 per hour, but the automation potential of this group is similar to that of people earning $90 to $105 per hour, for example. Similarly, higher education attainment levels and work experience are correlated with lower technical automation potential, but there is also a great deal of variability.

46. Using a linear model, we find the correlation between wages and automatability (the percentage of time spent on activities that can be automated by adapting currently demonstrated technology) in the US economy to be significant (p-value < 0.01), but with a high degree of variability (r2 = 0.19).

Technical potential is an important consideration, but other factors including cost of automation and wage levels also weigh on the decision to automate

The deployment of automation technologies in the workplace depends on a range of factors, of which the availability of the technology itself is an important one, but not the only one. That it is technically possible for a robot to carry out an activity does not mean that it will necessarily be deployed to do so in the workplace. Four other factors also need to be taken into consideration. The pace at which activities are automated and the extent of that automation reflects a subtle interplay between these factors and the trade-offs among them.

The first factor is the cost. Buying, adapting, and integrating the necessary hardware and software to automate activities can be expensive and complex, and before doing so, employers need a strong business case. Will machines in fact be able to undertake the tasks at hand less expensively and more efficiently, or do human labor and skills still have the edge? While it may be possible to automate service at a fast-food restaurant, the cost of the machines compared to the cost of humans earning a minimum wage will need to be calculated and considered. Indeed, a more clear-cut case for automation may come as higher-wage jobs are reviewed for their technical feasibility to be automated.

The cost of labor and the related supply and demand dynamics may thus play a significant role in decisions about automation, and are the second factor. If workers are in abundant supply and significantly less expensive than automation, this could be a decisive argument against automation. We calculate that just over $1 trillion in wages could be economically automated with a technology cost of $20 per hour, and $2 trillion could be captured with an automation cost of $10 per hour. Our analysis at the level of individual activities supports the argument that some occupations in the middle of the income and skill distribution are more susceptible to automation than others at the top and bottom (see Box 2, “Labor market polarization and the technical automation potential of occupation families).

Box 2. Labor market polarization and the technical automation potential of occupation families

A body of work in the economics literature documents “polarization” in the labor markets of developed
countries, in which “wage gains went disproportionately to those at the top and at the bottom of the income and skill distribution, not to those in the middle.” 1 These observations have largely been based on workforce data from national statistical agencies. Our analysis at the level of individual activities supports the argument that some occupations in the middle of the income and skill distribution are more susceptible to automation than others at the top and bottom. As depicted in Exhibit 7, occupation families (an aggregation of individual occupations) for transportation, office administration, and production, in the middle of the distribution, have higher percentages of activities with high technical
automation potential (collecting data, processing data, and predictable physical activities) than other occupation families lower and higher in the distribution. However, one occupation family at the low end of the income distribution, food preparation, has the highest percentage of time in activities with a high technical potential for automation. Furthermore, as technology continues to develop over time, the automation potential of different activities will also increase.

1. David H. Autor, “Why are there still so many jobs? The history and future of workplace automation,” Journal of Economic Perspectives, volume 29, number 3, 2015.

Exhibit 7

Exhibit 7

1. Data for the United States only.
2. Aggregations of individual occupations. Eight occupation families selected from a total of 22. NOTE: Numbers may not sum due to rounding.

SOURCE: McKinsey Global Institute analysis

A third factor to consider are the benefits of automation beyond labor substitution. These can include higher levels of output, raised quality, speed, agility, safety, and fewer errors. The potential savings from these benefits can be larger than those from labor costs. Finally, there is the issue of social and regulatory acceptability of automation. While selfdriving autos and trucks are undergoing tests in both the United States and Europe, they will be able to operate without human co-drivers only when regulators are comfortable with them doing so. Social acceptance may be even more difficult. While a robot in theory could carry out some functions of a nurse or a home-care help, the human beings on the receiving end of their care may balk at the idea.

Differentiating work activities into seven high-level categories

Across the more than 2,000 work activities across the US economy that we analyzed using Bureau of Labor Statistics data, we identified seven high-level categories of work activity. Each of these categories has a different potential for automation. Three categories have the highest technical potential for automation: performing physical activity and operating machinery in predictable environments, processing data, and collecting data. The other four high-level categories have a considerably lower potential for automation: performing physical activities and operating machinery in unpredictable environments; interfacing with stakeholders; applying expertise to decision making, planning, and creative tasks; and, least susceptible to automation, managing and developing people (Exhibit 8).

Exhibit 8

Three categories of work activities have significantly higher technical automation potential.

Exhibit 8

1. Managing and developing people.
2. Applying expertise to decision making, planning, and creative tasks.
3. Interfacing with stakeholders.
4. Performing physical activities and operating machinery in unpredictable environments.
5. Performing physical activities and operating machinery in predictable environments.
NOTE: Numbers may not sum due to rounding.

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

Performing physical activities and operating machinery in predictable environments

Almost one-fifth of the time spent in US workplaces involves performing physical activities and operating machinery in predictable environments, which have the highest automation potential of our seven categories, 81 percent. Currently demonstrated technology works best in these environments, where changes are relatively easy to anticipate. Physical activities in predictable environments figure prominently in sectors such as manufacturing, accommodation and food services, and retailing. That makes these sectors among the most susceptible to automation. Exhibit 9 shows a breakdown of different sectors of the economy based on the seven high-level categories.

In manufacturing, for example, performing physical activities or operating machinery in a predictable environment represents one-third of the workers’ overall time. The activities range from packaging products to loading materials on production equipment to welding to maintaining equipment. Because of the prevalence of such predictable physical work, almost 60 percent of all manufacturing activities could be automated. The overall potential, however, masks considerable variance. Within manufacturing, welders, cutters, solderers, and brazers, have an automation potential above 90 percent, for example, while that of customer service representatives is below 30 percent.

Manufacturing is the second most readily automatable sector in the US economy. A service sector occupies the top spot: accommodation and food services, where almost half of all labor time involves physical activities in predictable environments and the operation of machinery—including preparing, cooking, or serving food; cleaning food preparation areas; and preparing hot and cold beverages. According to our analysis, 73 percent of the activities workers perform in the accommodation and food services sector have the technical potential for automation.

Some of this potential is familiar. Automats, or automated cafeterias, for example, have long been in use. Now restaurants are testing new, more sophisticated concepts, such as self-service ordering or even robotic servers. Solutions such as Momentum Machines’ hamburger-cooking robot, which can reportedly assemble and cook 400 burgers an hour, could automate a number of cooking and food preparation activities. 47

Data processing and data collection

Data processing is the second category most readily automatable (69 percent) and accounts for 16 percent of all the time spent working in the United States. That is followed by data collection (64 percent automation potential and 17 percent of time spent). These activities are common to almost all sectors, ranging from human resources staff recording personnel history to mortgage brokers filling in forms, medical staff compiling patient records, and accounting staff processing payments. These are not just entry-level or low wage jobs; people whose annual incomes exceed $200,000 spend some 31 percent of their time doing those things as well.

47. Melia Robinson, “This robot-powered burger joint could put fast food workers out of a job,” Business Insider UK, June 30, 2016.

Exhibit 9

Technical potential for automation across sectors varies depending on mix of activity types.

Exhibit 9

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

Companies have long automated activities such as administering procurement, processing payrolls, calculating material-resource needs, generating invoices, and using bar codes to track flows of materials. But as technology progresses, computers are helping increase the scale and quality of these activities. For example, a number of companies now offer solutions that automate entering paper and PDF invoices into computer systems or even processing loan applications. “Robotic process automation” systems use software to automate well-defined data transactions currently performed by many workers.

Financial services and insurance provide one example of this phenomenon. The world of finance relies on knowledge work, not physical work: stock traders and investment bankers live off their wits. Yet about 50 percent of the overall time of the workforce in finance and insurance is devoted to collecting and processing data, where the potential for automation is high. Insurance sales agents spend considerable time gathering customer or product information, as do underwriters on verifying the accuracy of records. Securities and financial sales agents prepare sales or other contracts. Bank tellers verify the accuracy of financial data. As a result, the financial and insurance sector has the potential to automate activities taking up 43 percent of its workers’ time.

Performing physical activities and operating machinery in unpredictable environments

In creating our high-level groupings of activities, we divided the performance of physical activities and operation of machinery into two distinct categories depending on the environment or setting in which the activity takes place. As we have seen, carrying out this type of physical activity in a predictable setting has a rote quality to it, and the technical potential to automate is accordingly high, at 81 percent. When the environment or setting is unpredictable, however, the automation potential is much lower, at 26 percent.

Physical activities in an unpredictable environment make up a high proportion of the work in sectors such as forestry and construction. Examples include operating a crane on a construction site, providing medical care as a first responder, collecting trash in public areas, setting up classroom materials and equipment, and making beds in hotel rooms. The environment in these examples is not stable and can change in unpredictable ways. Carrying out these activities thus requires a high degree of flexibility, which makes them harder to automate. For example, people in public areas do not always drop trash in the same place, and while sometimes they might drop paper, at other times they leave plastic bottles or soda cans.

Interfacing with stakeholders

Across sectors of the economy, especially in services, workers interface on a regular basis with a wide range of stakeholders—customers, patrons, or visitors. Greeting them, for example in a retail store, explaining technical product details or service information to customers, including from a call center, responding to complaints and questions, scheduling appointments, or providing advice are just some of the numerous types of interactions that take can place with stakeholders. These types of activity for now have a relatively low potential for automation based on currently demonstrated technologies, just 20 percent. Social and emotional responses are important for these tasks, as are linguistic and cognitive capabilities such as logical reasoning and problem solving.

Applying expertise to decision making, planning, and creative tasks

Even where computers perform above human levels in some well-defined activities such as optimizing trucking routes, humans—for now—still need to determine the goals, interpret the results, or provide commonsense checks for solutions. Activities that require application of decision making, planning, and creativity account for 14 percent, about one-seventh, of the total time spent working in the United States, but they have relatively low automation potential, at just 18 percent, based on adapting currently demonstrated technologies. These activities can be as varied as coding software or creating a menu, developing marketing plans, or writing promotional materials. They are common in fields such as education, human resources, and finance, and considerable amounts of time spent on them in the US economy involve evaluating students’ work, coordinating operational activities, and examining financial records or processes.

Managing and developing others

The category of activities we describe as managing and developing others has the lowest automation potential. Only about 7 percent of time in the workplace is spent on these activities, and the potential to automate them is low, about 9 percent. Chief executive officers and senior managers spend a significant proportion of their time engaged in such activities, which brings down their overall automation potential; about 25 percent of a CEO’s daily activities could be automated using currently demonstrated technology, but that mainly represents data collection and analysis, rather than talent management.

Among the sectors, education is among the least susceptible to automation, at least for now, with an automation potential of 27 percent. To be sure, digital technology is transforming the field, as can be seen from the myriad classes and learning vehicles available online. Yet the essence of teaching is deep expertise and experience, and complex interactions with other people. Together, those two categories—the least automatable of the seven identified in Exhibit 6—account for about half of the activities in the education sector.


We began our research into automation potential by focusing on sectors of the US economy, to establish the methodological framework that underpins our research. We then broadened the analysis to a total of 46 countries, using comparable national and international data, where available, including wage data from foreign direct investment sources. Details of our international methodology are in the technical appendix.

Overall, currently demonstrated automation technology has the potential to affect activities associated with 40 to 55 percent of global wages depending on country (Exhibit 10). This amounts to about $15.8 trillion in wages and the equivalent of 1.1 billion workers. As we will see in Chapter 4, the actual deployment of automation could vary widely from country to country, depending on a number of factors including the level of wages and the cost of deploying solutions.

The key differences in the total wages associated with technical automation potential among countries results from differences in the mix of sectors within each economy, the mix of occupations within sectors, and wage levels.

In terms of total wages associated with technically automatable activities, the potential is concentrated globally in China, India, Japan, the United States, and the five largest European Union countries—France, Germany, Italy, Spain, and the United Kingdom. These are the countries with a combination of the largest labor forces or higher wages.

Exhibit 10

The technical automation potential of the global economy is significant, although there is some variation among countries.

Exhibit 10

1. Pakistan, Bangladesh, Vietnam, and Iran are largest countries by population not included.

SOURCE: Oxford Economic Forecasts; Emsi database; US Bureau of Labor Statistics; McKinsey Global Institute analysis

Automation potential is concentrated in China, India, Japan, the United States, and the largest European Union nations

More than half the wages and almost two-thirds of the total number of workers associated with technically automatable activities are in just four countries—China, India, Japan, and the United States. These four together account for about $9 trillion of the wages and more than 700 million employees of the global total potentially affected. In the five largest European Union economies—France, Germany, Italy, Spain, and the United Kingdom— more than 50 million workers and $1.7 trillion in wages are associated with technically automatable activities (Exhibit 11).

The largest amount of employment associated with technically automatable activities is in China and India, because of the relative sizes of their labor force. Technically automatable activities make up the equivalent of more than 600 million full-time workers in the two countries together. In terms of wages associated with technically automatable activities, however, the United States is closer to China’s level ($2.7 trillion in the United States vs. $4.1 trillion in China) because of higher wage levels.

Exhibit 11

Technical automation potential is concentrated in countries with the largest populations and/or high wages.

Exhibit 11

1. France, Germany, Italy, Spain, and the United Kingdom.
NOTE: Numbers may not sum due to rounding.

SOURCE: Oxford Economic Forecasts; Emsi database; US Bureau of Labor Statistics; McKinsey Global Institute analysis

Differences and similarities in automation potential globally

Our analysis of the technical automation potential of the global economy shows that there is a range among countries of about 15 percentage points. Two factors explain this range. The first is the sectoral makeup of each economy. That is, the proportion of a national economy that is in sectors such as manufacturing or accommodation and food services, which both have relatively high automation potential, compared with the proportion in sectors with lower automation potential, such as education, management, and health care. The second factor is the occupational makeup of sectors in different countries. That is, to what extent workers in these sectors are engaged in job titles with high automation potential such as manufacturing production, and those in job titles with a lower automation potential such as management and administration. This weighting changes the automation potential of a sector depending on the country.

Two examples illustrate these differences. The first is China and India (Exhibit 12). With the world’s largest workforces, they have similar automation potential and dynamics overall: both have technical automation potential of 50 percent. They have similar top sectors, including agriculture, manufacturing, retail, construction, and transportation and warehousing, and the automation potential within each sector is very similar. Manufacturing and retail play a larger role in China than India, whereas agriculture accounts for a significantly greater share of hours worked than in India than in China as a percentage of the total. Within the sectors, the essential differences result from varying types of job families. For example, India has more welders and sewing machine operators engaged in manufacturing production than China, and both of these job families have a higher automation potential than many other types of jobs, such as managing and developing people, and specialized expert technicians. At the same time, India has a lower proportion of jobs requiring interactions with stakeholders and managing and developing people, activities with low automation potential.

Exhibit 12

Exhibit 12

NOTE: Numbers may not sum due to rounding.

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

The second example highlights differences between Japan and the United States (Exhibit 13). Japan overall has an automation potential of 55 percent of hours worked, compared with 46 percent in the United States. This difference is primarily due to a different sectoral mix in the two economies, and within those sectors, a different weighting of jobs with larger or smaller automation potential. For example, the automation potential of Japan’s manufacturing sector is particularly high, at 71 percent (compared with 60 percent in the United States). Japanese manufacturing has a slightly larger concentration of work hours in production jobs (54 percent of hours vs. 50 percent) and office and administrative support jobs (16 percent vs. 9 percent). Both of these job titles comprise activities with a relatively high automation potential. By comparison, the United States has a higher proportion of work hours in management, architecture, and engineering jobs, which have a lower automation potential since they require application of specific expertise such as high-value engineering, which computers and robots currently are not able to do. These differences outweigh the higher level of wages in the United States than Japan, which affect the business case for automation.

Similar differences exist among countries globally, for example, between Argentina and Brazil, France and Germany, or Kenya, Nigeria, and South Africa. A detailed look at all 46 countries we have examined is available online. 48

48. The data visualization can be found on the McKinsey Global Institute public site at!/

Exhibit 13

NOTE: Numbers may not sum due to rounding.

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

The technical potential to automate work activities in the global economy by adapting currently demonstrated technologies is already close to 50 percent, thanks to rapid advances in automation technologies. That does not mean automation will occur overnight, however, since technical potential is only one of several factors that will eventually lead to automation adoption in the workplace. Only a tiny fraction of entire occupations could be automated with current technologies, but activities across a wide spectrum are susceptible, especially those involving predictable physical activities, data processing, and data collection. What is the outlook for the development of these capabilities? How rapidly could technical potential become workplace adoption? And what would the workplace then look like? In the following chapters, we model some ways automation could arrive in the workplace and lay out timeline scenarios for its actual adoption across the world.