Report from the McKinsey Technical Institute

Our analysis of automation potential inevitably raises questions about the practical effect on workplaces and on the people who work in them. What exactly will change, and how? To develop a vision of automation as it could be applied in the workplace, we have created hypothetical case studies across different industries that suggest how automation could affect specific processes. In this chapter, we outline five of these cases by way of example. The five are from highly varied sectors and involve a broad range of activities across various industry sectors and physical settings: a hospital emergency department, aircraft maintenance, an oil and gas control room, a grocery store, and mortgage brokering.

For all their differences, our case studies also have several essential elements in common. First is the changing nature of work itself, which will likely affect all workers at all skill levels. There will be less routine and repetitive work based on rules-based activities, because this can be automated across many occupations and industries. This in turn will mean many workers may need to acquire new skills. The workplace will become ever more a place where humans and technology interact productively. A major part of most human jobs will involve working together with artificial intelligence, robotics, and other technologies. Automation will affect more than distinct work activities: processes and procedures will also likely have to adapt, too. This in turn will have profound implications for how the workplace is structured and organized.

A second essential element of automation in the workplace will be its impact on business economics. Advances in automation technologies are often depicted simplistically as robots replacing humans. In fact, in evaluating the economics of our case studies, we see that automation’s effects are twofold. While it will result in some degree of labor substitution in all five of our case studies, it will also drive significant performance and quality gains in four of the five and, in one of these cases—the oil and gas control room—these performance improvements far outweigh the gains from labor substitution. These improvements are the result of automated systems carrying out a range of activities better than human workers, and include eliminating waste, improving efficiencies, heightening safety, and enhancing quality. As a rule of thumb, capital-intensive industries tend to accrue automation benefits mostly through performance gains, as capital is better utilized, while labor-intensive sectors tend to benefit more from labor substitution. Together they provide a strong rationale for the deployment of automation technology.

The third essential element is that the business case for automation is often strong. The relative cost of automation is likely to be modest compared with the value it can create. The types and sizes of investment needed to automate will differ by industry and sector. For example, industries with high capital intensity that require substantial hardware solutions to automate and are subject to heavy safety regulation will see longer lags between the time of investment and the benefits than sectors where automation will be mostly software-based and less capital-intensive. For the former, this will mean a longer journey to breakeven on automation investment. However, our analysis suggests that the business case is compelling regardless of the degree of capital intensity: the run-rate benefits of investment in automation in our case studies are between three and 11 times the costs of that investment.

These case studies are not precise projections. The vision of the future they provide is based on the hypotheses of industry experts about how these processes could be transformed by automation. While these case studies are partly anecdotal, we nonetheless believe they are a useful exercise that could indicate key benefits and challenges related to developing and deploying automation technologies in various sectors, as well as providing a vision of how automation could actually transform the workplace.

For all five cases, we outline two scenarios: an “interim” future state as automation technology is deployed in current processes and structures, and a “provocative” future state in which we envisage automation being used in a changed structure or process environment that is tailored to maximize its advantages. We also outline some of the likely barriers to automation deployment in each of the cases, including legal and policy obstacles, organizational impediments to change, difficulties in integrating technology, economic viability, and human reluctance to accept technology solutions in certain situations.


Two common characteristics of many hospital emergency departments today are their high level of human interaction and long patient waiting times; in the United States it is rare to be discharged in less than two hours on average. 49 Automation has the potential to reduce those waits and increase productivity, as doctors and nurses focus more effectively on better outcomes, and machines take on routine activities such as registration, checkout, and dispensing of prescriptions (see illustration, “Hypothetical future state of a highly automated emergency department”). Predictive health care using sensing wearables to check vital medical signs and remote diagnostics could cut patient waiting times. For hospitals, automation could streamline billing and other administrative activities.

To achieve such outcomes, hospitals will need to make significant investment in automation technology, along with time and capital to train staff. They will also need to redesign process workflow. Doctors and nurses will have to become comfortable working closely with and trusting automated systems. Safety and liability are significant challenges in a sector where malpractice suits are common; in the United States, about $3.6 billion was paid out in malpractice suits in 2014, and artificial intelligence companies could find themselves on the receiving end. 50 Stringent privacy regulations will need to be safeguarded. The emergency room is also a place where human unease with machines could be strong: people who come to hospitals for medical emergencies usually want and need to interact with qualified humans and may not feel comfortable with machines, however medically competent. 51

In emergency departments in the United States today, about 80 to 85 percent of the patients are walk-ins, and about the same percentage of the total are treated and sent home usually with prescribed medicines. 52 Patients interact with a range of workers. First are the medical secretaries who enter and validate patient information. Triage nurses check vital signs, request laboratory tests or imaging, and decide to discharge or refer a patient to see a doctor. Doctors examining a patient prescribe medicines and decide to discharge, admit to hospital, or refer the patient to a specialist. Lab technicians conduct tests. At the end of the process, medical secretaries collect payment or compile documentation for an insurance claim.

At least some of these activities could be fully or partially automated. They include the initial work of gathering patients’ information, checking vital signs and requesting lab reports. Lab registration and tests are also potentially automatable, as is the end process of payment. Some aspects of a doctor’s work in an emergency department are also potentially automatable—not just the data collection that takes up some of a doctor’s time, but also some areas of disease diagnosis, and even some aspects of medical procedures and surgery. For example, in radiology, computers are already analyzing X-rays, CT scans, and MRI imagery. Completely automated diagnosis is not likely to happen quickly, partly for reasons of patient acceptance, and partly because of the technical difficulty of integrating data from multiple sources (including natural language understanding, recognizing and processing the patient’s emotions) to determine a diagnosis and course of treatment. Automated diagnostic advice is thus likely to augment doctors’ decision making before fully automated diagnosis, except perhaps in special instances, such as radiology, or cell pathology (checking for abnormalities through a microscope).

Overall, we calculate that about 30 percent of the benefits of automation in an emergency room would come from performance gains, and 70 percent from labor substitution. Productivity could rise substantially, while the number of full-time equivalents could decline by about half, with the main reductions at the registration desk and in lab testing.

Case Study 1

49. ProPublica lists average waiting and treatment times in US emergency rooms, broken down by state, at See also Lisa Esposito, “Enduring really long waits at the emergency room,” US News and World Report, May 8, 2015.
50. 2015 Medical malpractice payout analysis, Diederich Healthcare, May 2015.
51. The unsettling feeling that people may experience when humanoid robots or audiovisual simulations closely resemble humans is known as “uncanny valley,” a term coined by a Japanese robotics professor, Masahiro Mori, in a 1970 essay. The eeriness is largely prompted by the machine’s appearance, which is realistic but not convincingly so.
52. Cheryl Gutherz and Shira Baron, “Why patients with primary care physicians use the emergency department for non-emergency care,” Einstein Journal of Biology and Medicine, volume 18, number 4, 2001


Maintaining a commercial aircraft takes an average of about 8,000 technician hours per plane annually. 53 Technicians conduct visual inspections of the aircraft for signs of physical wear and damage. They also remove and replace parts, conduct a sign-off for quality, and handle administrative tasks, including supplying records to the Federal Aviation Administration where needed. These are jobs that involve spending considerable amounts of time on walking around the aircraft and waiting on parts, planes, or people. They are also dangerous: in 2013, 57 job-related deaths in the United States were in aircraft maintenance. 54

Automation could have a major beneficial impact on the sector (see illustration, “Hypothetical future state of highly automated aircraft maintenance”). First, removing technicians from fall hazards and fuel tanks would represent a significant improvement in safety. Second, robots equipped with image process algorithms already do a better job than humans at image identification, and deploying them would improve the detection of defects in the aircraft; maintenance errors caused more than one third of the 179 commercial jet engine accidents between 1988 and 2013. 55 Automated warehouse systems could eliminate about 75 percent of the time wasted by walking around the aircraft, picking up tools and parts,
and improved sensors and analytics could raise the proportion of predictive maintenance, which is less costly than reactive maintenance. For maintenance companies, the savings from reduced waste and a move to on demand maintenance would save costs. Automation could enable experts to monitor all maintenance from a command center, which would reduce variability, and ensure better data collection. Consumers would benefit from these savings if they are passed on in the form of lower flight costs. About half of engine-related delays today are caused by maintenance issues, and so flight delays could also be reduced. 56

For technicians themselves, automation could change the workplace and their roles. Much of the routine work they carry out, including walking around the plane, moving it, and logging work records, could be automated. Already in an interim phase, remotely controlled robots could crawl through planes and inspect fuel tanks. High-resolution cameras guided by experts could inspect exteriors. Artificial intelligence algorithms could suggest potential problems based on logs even before inspection takes place. In our more advanced automation scenario, small robots could inspect the airframe without moving panels. Automated tugs would move planes, while robotic carts bring and remove parts and tools based on scheduling routines. These and other changes would allow technicians to be more focused on knowledge and handling exceptions, which require greater training.

Overall, we estimate that 35 percent of the value created by automation in aircraft maintenance could come from performance gains, while 65 percent could potentially come from labor substitution.

Case Study 2

53. Stephen Holloway, Straight and level: Practical airline economics, Routledge, 1997.
54. US Bureau of Labor Statistics, 2015.
55. Mary S. Reveley et al., Causal factors and adverse events of aviation accidents and incidents related to integrated vehicle health management, National Aeronautics and Space Administration, March 2011.
56. Jeffrey O’Brien, “Maintenance: What causes aircraft accidents?” Fiix, May 9, 2012.


Most of the gains of automation in oil and gas control rooms are likely to come from higher productivity and safety (see illustration, “Hypothetical future state of automation in oil and gas operations”). A control room is an operations center that monitors and controls upstream exploration and drilling operations. For now, key roles in the oil and gas sector are divided between onshore and offshore facilities, with operations and maintenance on an offshore rig largely overseen and carried out by managers and workers on the rig, while teams of petroleum and other engineers provide technical support and coordinate activities from headquarters onshore. Offshore work can be a dangerous business: the job-related fatality rate of people working in extraction operations in 2013 was four times the average for the US economy. 57 The performance and training of offshore operators vary, and errors are sometimes made, including ones with severe consequences.

Automation technologies can considerably raise the performance of control rooms by removing operators from environments that are hazardous and expensive to maintain and by capturing data that can be used for predictive and preventive maintenance and operational best practice. Centralizing expertise in an onshore location can improve strategy planning and the effectiveness of event responses. In our more futuristic scenario, permanent seafloor robots could undertake repairs or conduct additive manufacturing, while 3D printers on a surface vessel could print out replacement parts as needed. Robots would conduct standard maintenance, and operating algorithms developed using historical logs could deliver more efficient operations and greater safety.

The advantages of automation in this case include better personnel safety, greater efficiency, higher throughput, improved agility, and cost reductions from relocating operators from remote sites to centralized offices. Improved sensors for remote operations and analytics can enable predictive maintenance, which is just one quarter the cost of reactive maintenance. 58

Overall, we estimate that 80 percent of the value created from automation in oil and gas control rooms would come from performance gains, with the rest from labor substitution. To reap such gains will require the integration of technologies so that cognitive, sensory perception, and physical capabilities can be deployed on site. Data scientists, engineers, and developers will be needed to develop algorithms that optimize remote management.

Case Study 3

57. Occupational fatalities during the oil and gas boom—United States, 2003–2013, Morbidity and Mortality Weekly Report, Centers for Disease Control and Prevention, May 29, 2015.
58. Elly Earls, “Staying one step ahead with predictive maintenance technology,” Mining, October 15, 2013.


Imagine walking into a grocery store and being greeted by name, courtesy of facial recognition software. Thus begins a highly personalized shopping experience that is faster, more tailored to your preferences, and more convenient than shopping today. If you cannot find what you are looking for, you provide instant feedback. You might use personalized coupons on your mobile device. Once you have finished your selection, an automated back-room service sends out your goods, or a drone drops them off at your home. Best of all, there is no line for payment because there is no physical checkout. The store senses what goods you have with you when you walk out of the store, and payment is accurately and automatically deducted from your preferred account (see illustration, “Hypothetical future state of automation in a grocery store”).

We are not all that far from such a scenario today. Stores still have physical shelves, but self-checkout kiosks are becoming commonplace and are a short step from automatic payment and shipping. 59 Robot cleaners and automated storerooms already exist. Augmented virtual reality product views are just a question of time, where consumers will be able to look at displays of goods (probably using special glasses or other technology) to find out more information about the product such as ingredients and nutritional details. One of the biggest overall benefits of automation in retail will be the improved customer experience, as the online and offline shopping experience merges into one, even as it becomes more individualized.

From the employment perspective, our case study suggests that automation in grocery stores could have a significant impact on staffing needs, with a reduction of about 65 percent of hours, mainly for front-end cashiers and people engaged in stocking and cleaning. Some workers could be repurposed toward higher value-added activities such as customer engagement.

The retail sector overall will face some significant changes. Space productivity will rise, and that in turn will reduce the need for large stores. Smaller stores require less investment; we estimate savings of 60 to 80 percent. Lower inventory and working capital will also be a feature of the new retail business landscape, as the brick-and-click models merge into a seamless whole, and retailers leverage physical stores as distribution centers. Data analytics will enable retailers and manufacturers to customize and target promotions.

Some customers may find marketing use of their personal data, including location-based alerts, overly intrusive. Many jobs in retail are entry-level, low-skill ones, and the elimination of many such positions could cause a public outcry. Training will be essential so that staff can troubleshoot technical and other problems and make recommendations. From a technology perspective, sensory perception will need to be integrated with pattern matching, so that customers are recognized and given relevant recommendations.

While the potential savings from smaller formats, less inventory, and lower payroll could be significant, the slim margins in retailing may mean that only large chains will have the capital needed to make the investment in full automation. Overall, we estimate that the benefits of automation will be three times the cost. Along with health care, this is the lowest ratio of our five case studies. Labor substitution gains in retail accounts for 68 percent of the potential gains, compared with 32 percent from improved performance.

Case Study 4

59. Amazon in December 2016 launched a store without cashiers, Amazon Go. Leena Rao, “Amazon Go debuts as a new grocery store without checkout lines,” Fortune, December 5, 2016.


The financial-service industry has been at the forefront of adopting automation technologies for a range of back-office work. This case shows how that degree of automation could reach even higher levels. It focuses on the business of processing mortgage applications, which consists in large part of collecting and processing data, two categories of activity that already have a high automation potential, by adapting currently demonstrated technologies. Our scenarios for automation in this process accordingly correspond to a sharp reduction in the overall required labor hours, of between 55 and 85 percent. The speed with which mortgages are processed could accelerate substantially. For now, it takes an average of 37 days to approve a mortgage application in the United States, of which about 14 to 21 days are spent on the mechanics of application processing. 60 In our interim automation scenario, in which technology is deployed in current processes and structures, this could drop to less than six days. In a more futuristic scenario, in which the processes themselves change, mortgage approvals could come through in less than a day. The shorter turnaround time could improve the dropout rate by possibly 30 percent or more, as many people who apply for mortgages today drop out as a result of the lengthy process.

For mortgage companies, the improvements derived from automated processes could lower default risks and eliminate inconsistencies in processing, thereby reducing the need for human performance management. Automation could also create a potential for partnership between mortgage companies and real estate agents for real-time lead generation, mortgage application, and loan fulfillment. Customer satisfaction could rise as a result of instant pre-approval, hassle-free applications, and much faster turnaround. Moreover, some of the industry’s challenges, including human bias in underwriting and the difficulty that underserved borrowers can have in accessing affordable capital, could also be eased by automated processes. Of our five cases, however, the automation of mortgage applications will potentially weigh most heavily on labor. The vast majority of the impact in this process, about 88 percent, would come from labor substitution gains, compared with just 12 percent from performance gains. Moreover, the business case to move to automation is strong, since the software costs could be relatively low, and wages of loan officers are relatively high, at $35 per hour. 61 The run-rate benefit from automated mortgage origination could be as much as 11 times the cost of automation itself, according to our estimates, easily the highest ratio of our five case studies.

For mortgage brokers, automating and speeding up the approval processes will allow more time on complex tasks such as advising customers or handling exceptions that require human expertise and judgment, as well as additional time to manage unusual applications. That in turn will require more training. Data scientists will also be needed to increase the accuracy of algorithms used in processing, and to integrate platforms and data sources. Nonetheless, there is likely to be considerable restructuring and elimination of redundant positions in the core application processing functions, which could require managing labor issues, including helping to redeploy displaced workers.

To reap the full benefits of automation, the entire mortgage approval process will need to be redesigned and piloted in select branches to ensure all necessary resources are available. There are technological challenges in ensuring end-to-end integration with systems from retail banks and underwriters, such as handling documents, messaging, and conducting risk analysis. However, given the potential cost savings and fairly certain return, financial institutions have an incentive to invest in automation for their mortgage origination activities, much as they have for straight-through processing of other types of transactions.

60. Evan Nemeroff, “Survey shows average mortgage closing time hits three-year low,” National Mortgage News, August 22, 2014.
61. Occupational Employment Statistics, US Bureau of Labor Statistics, 2014.

No sector of the economy will be immune to automation. While our five hypothetical case studies are not precise predictions about how automation will be deployed in the workplace, they do signal some of the key benefits and challenges to come. Integrating technological capabilities, retraining staff, adjusting processes, and, in some sectors, working out how to make workers and customers comfortable interacting with machines rather than humans, are among the considerable changes that will be required to maximize the benefits from automation. Gains from labor substitution could be significant, but they will vary from sector to sector. Performance gains, in the form of better quality, greater safety, and higher productivity, could also be substantial, to judge by the findings of our case studies. How quickly will any of this happen? In the next chapter, we analyze the factors that will hasten or slow the adoption of automation, and project timelines for its implementation in the global economy.