Idea in Brief
Despite advances in automation technology, the promise of productive and flexible automation, with minimal involvement of human workers, is far from reality.
The adoption of automation technology has been limited. And when firms do automate, what they gain in productivity they tend to lose in process flexibility, a zero-sum outcome.
Positive-sum automation measures success across three levels: the machine, the system, and the team. You’ll know you’ve succeeded when the automation makes your human teams happier and better at their jobs.
In 1982, General Motors announced it was building a “factory of the future.” The Saginaw, Michigan, facility would automate production, revitalizing GM’s business at a time of intense competition from Japanese automakers Toyota and Nissan. GM had posted a loss of $763 million two years earlier—only the second losing year in its 72-year history. When CEO Roger Smith returned from visiting a Toyota factory, he resolved that GM must automate to compete.
The Saginaw project envisioned an army of 4,000 robots running production. The goal was to increase productivity and flexibility. The robots would slash up to two years from GM’s five-year production cycle and be capable of switching between diverse GM models. Employee productivity would increase 300%. Manual systems and interfaces would be eliminated. The robots would be so effective that people would be scarce—it wouldn’t even be necessary to turn on the lights.
But GM’s “lights out” experiment was a mess. Production costs in the factory of the future exceeded those in plants employing thousands of unionized workers. In several facilities, the robots struggled to distinguish one car model from another: They tried to affix Buick bumpers to Cadillacs, and vice versa. The robots were bad painters, too; they spray-painted one another rather than the cars coming down the line. GM shut the Saginaw plant in 1992.
In the three decades since the plant’s closure, scientists and engineers have made remarkable advances in robotics hardware (the physical machines) and automation software (the computing intelligence powering the machines). Robots and other automation technology perform repetitive tasks with increasing safety and accuracy. They can cut and weld metal consistently and without injury. They can paint cars without painting one another. And automation now has applications in new and more-sophisticated contexts beyond the factory floor.
Despite advances in automation technology, however, the promise of lights-out manufacturing—productive and flexible automation with a minimal number of human workers—is far from reality, for two main reasons. First, adoption of the technology has been halting and limited. According to 2018 U.S. Census data, fewer than 10% of U.S. manufacturing firms reported using robots. In 2020, when the Covid pandemic and stay-at-home orders were expected to increase demand for factory automation, robot purchases in the United States, Germany, and Japan fell below 2019 levels. In China, despite heavy subsidies for robot adoption as part of a national strategy to drive automation, the share of manufacturers using robots is estimated to be roughly the same as in the United States. And even when firms do adopt automation technology, studies show, they end up hiring more workers, not fewer, as they become more productive.
The costs of switching over an automated system to do something new are frequently much higher than switching over a team of human workers.
Second, our research shows that what a company gains from automation in productivity it tends to lose in process flexibility. Routine maintenance on a robot (to recalibrate sensors, for example) can grind production to a halt while third-party consultants are called in. Preprogrammed robots are locked into rigid ways of accomplishing tasks, stunting innovation by line employees. And so on. We call this zero-sum automation.
Drawing on our experience researching, developing, and deploying AI and robotics, along with dozens of interviews and site visits conducted as part of MIT’s Work of the Future task force, we’ve found that companies can avoid zero-sum automation—if they abandon the lights-out playbook. They must stop measuring project success by comparing the cost and output of machines with the cost and output of human workers; that approach overlooks how automation can contribute to improving a process across multiple dimensions. Instead, companies should focus on questions like: Will the team that currently performs the tasks to be automated be more productive doing something new? Will teams using automation technology generate more-innovative ideas or take on more-varied tasks than teams without it?
In this article, we introduce the concept of positive-sum automation, which we’ve defined as the design and deployment of new technologies that improve productivity and flexibility. Positive-sum automation depends on designing technology that makes it easier for line employees to train and debug robots; using a bottom-up approach to identifying what tasks should be automated; and choosing the right metrics for measuring success.
The Limitations of “Lights Out” Automation
Automation technologies that are designed to maximize productivity tend to limit flexibility in three key ways: 1) They are not readily adaptable to changes in their external environment; 2) they require specific, deeply technical skills to program and repair them; and 3) they tend to be “black boxes,” operating without human feedback or input. Those limitations often force companies to ditch the lights-out goal and rely instead on the flexibility, creativity, and improvisation skills of human workers.
Elon Musk tried to revive the idea of a lights-out factory in 2017 to mass-produce Tesla’s Model 3. The company built robots to help boost production in its California factory and overcome the challenges of hiring and training workers. But Tesla ran into production delays and struggled to navigate what Musk described as a “crazy, complex network of conveyor belts.” Like GM, Tesla reversed course, abandoning some of its investments in automation and scaling up its skilled workforce. “Humans are underrated,” Musk concluded.
In China, manufacturers have come to a similar conclusion. They originally planned to use robots widely across factories to manipulate and assemble electronic components, but it turned out that the robots couldn’t perform the delicate tasks required in electronics assembly as well as humans could. Harvard sociologist Ya-Wen Lei quotes one manufacturing executive as saying, “Robots often break delicate and expensive components. From the process, I have realized that the human body is magic.”
Or consider an example from outside the world of manufacturing and robotics. The MD Anderson Cancer Center enlisted IBM’s Watson in 2013 to help doctors quickly find treatment options within vast databases of research. But the software had difficulty making sense of patients’ complex medical records and needed extensive human input to offer diagnostic advice. In some cases, Watson surfaced evidence that was unreliable or incomplete. And when medical evidence changed—for instance, a new clinical trial suggested a new approach to treatment—humans needed to manually update Watson’s recommendations. After an initial wave of enthusiasm, users determined that Watson’s applications were limited. MD Anderson canceled the program in 2017.
When a robot’s external conditions change—which they inevitably do, as when a firm wants to update its production process or begin producing a new version of a product—the automated system needs to be reprogrammed, retested, and retaught. The costs of switching over an automated system to do something new are frequently much higher than switching over a team of human workers. One reason the switching costs are so high is that the expertise to adjust, repair, and reprogram the automated system typically comes from people outside the team that uses it. A production team might rely on a third-party integrator or repair team to reprogram an automated system. A hospital’s accounting team might need to call in IT to fix software when the billing system breaks. It’s at this point that the lights go out on “lights out.”
To achieve positive-sum automation, companies must design systems for both productivity and flexibility. We see three keys to automating flexibly.
Design easily comprehensible tools and invest in training.
Many robots and automated systems are designed and configured by third-party technical consultants in ways that make them rigid and brittle. Even small changes in the production environment or process can stymie the system. To avoid such issues, companies should make sure that automation systems incorporate easily comprehensible technology such as lower-code programming interfaces that enable line employees with little technical skill to repair or adjust them in real time.
Consider this example of workers’ declining to use automation because they couldn’t fine-tune the way it worked. In an American factory for assembling scientific sensing equipment, a robot works in close collaboration with a technician. When the technician presses a pedal, the robot maneuvers the assembly overhead, rotates it to the left, and tilts it down and forward, where the technician can perform the dexterous work of placing fasteners and installing delicate sensors. Together, the technician and the robot can complete the tasks in equal or less time than the technician can alone. The robot saves the technician from craning her neck or twisting her wrist into uncomfortable positions. But the robot often goes unused. When given a choice, technicians prefer the next station over, where they can perform the task without the robot’s help. When one worker was asked why, she said that the robot’s set of motions were preprogrammed, but she’d prefer to do the steps in a different sequence. Because the system is built so rigidly, with complex code underlying the robot’s movements, the technician can’t adjust the robot or her workspace according to her preferences.
Automations that can be flexibly tasked and directed by line employees enhance and accelerate innovation.
Start-ups and research labs are now focusing on low-code automation software that can assist a line employee in configuring and troubleshooting a robot. Other low-code tools empower robots to learn new multistep tasks from a human expert. The human demonstrates the process for the robot, which watches and learns. When it is ready to perform the task, the human observes the process to ensure that the robot is doing it properly.
In addition to choosing the right hardware and software, companies should invest in training to build line employees’ independence in not only operating the technology but also reconfiguring it for new applications. Training should encompass multiple people across multiple roles to ensure that there isn’t a single point of failure and that different perspectives to designing, integrating, and measuring outcomes are considered. Companies investing in automation need to stay current on how the technology is evolving and identify new opportunities to refine or beef up skills as it improves.
Solicit feedback from line employees.
When firms use a top-down approach to automation, the primary goal is often to maximize productivity. Senior managers analyze the organization’s processes, and with the help of a consulting firm or an IT team, they build the tools for automation. But senior leaders usually lack a detailed understanding of what the process entails, how much flexibility must be built into the automation, and what types of situations it might be unable to handle. A bottom-up approach puts line employees with the closest perspective on how a process is run in charge of recommending and developing how it is automated. Our research shows that automations that can be flexibly tasked and directed by line employees—a shop-floor worker, a billing specialist, a customer-service agent—enhance and accelerate the worker’s and the firm’s ability to innovate. And implementing automation from the bottom up makes it easier to win buy-in from workers.
Mass General Brigham has pursued a bottom-up approach to administrative automation throughout its hospital system. It started by hiring a consulting firm, which helped identify a suitable technology, and then asked the distributed teams in its administrative departments which tasks to automate. The employees close to the routine processes identified several mundane activities, such as tracking patient referrals to specialist clinics, checking that employee licenses are up to date, and managing incoming payments. The hospital then recruited individuals to learn how to program the bots, focusing on finding talent internally, particularly from teams that would be implementing the automation. The individual team members worked with those trained to program the bots to identify exactly how to match the software to the intricacies of the process. The people whose tasks were being automated supported the project, because the bots, which first went live in 2018, relieved them of work that they found especially mind-numbing.
Ohio-based G&T Manufacturing began a similar transformation in 2016. The 20-person factory produces a variety of parts for industries ranging from aerospace to agriculture. Its employees were once tasked with physically moving 40-pound machine parts into and out of a lathe that cuts and shapes the metal parts, repeating the process many times an hour. G&T wanted to automate that manual labor task. Companies in similar situations often rely on the expertise of a third-party integrator to help manage the automation process.
An integrator helped G&T get robots started, but G&T’s vice president, Colin Cutts, taught himself how to train and retrain them. He then taught G&T’s machinists to program the robots and troubleshoot problems. They developed libraries of programming instructions for the shop’s robots that can be adapted as G&T switches from producing one part to another, when it improves a process, or when it’s exploring something new. Cutts’s goal is to make the software skills—the specialized knowledge to adapt robots to a changing production environment—part of a machinist’s everyday work.
Before G&T adopted this new system, there was one machinist per machine, loading parts, unloading them, and inspecting them. Now there’s one machinist for every three machines, operating in a supervisory role. Rather than lifting and loading, machinists focus on inspecting parts and responding to problems as they arise. Since the task was automated, scrap and waste at G&T has dropped from 12% to less than 1%, and output per worker has more than tripled.
Choose the right KPIs.
It would be impossible to provide a single equation that can determine automation success. Companies should develop KPIs that consider each process to be automated, each team involved, and each employee whose tasks might change. They should also factor in intangible benefits, including product innovation, improved employee satisfaction and safety, and reimagined processes.
Productivity is the number one motivation for firms adopting automation technology, but when we dug deeper and asked managers to explain their decisions in more detail, we found that their motivations varied widely. Some companies built an automation to handle dangerous tasks. Some chose to automate tasks that their workers would rather not do. Others focused on waste reduction or improved process reliability. A few firms we spoke with had adopted robots out of curiosity or because their competitors were doing it; they were still figuring out the business case months after the implementation had started.
The challenge for businesses with nuanced motivations is that measuring success must also become nuanced. In some cases, an apples-to-apples comparison of a manual system with an automated one won’t make sense: Automated systems require process reengineering—removing steps that are inefficient and perhaps adding others. To account for this, companies should develop a range of metrics at three levels: the machine, the system, and the team. At the machine level, success measures might focus on practical flexibility: How long does it take for an automated system to learn a new task versus a human worker? At the system level, the measure might focus on switching costs: How long does it take a robot or automated software to get a new process up and running?
We consider the measures of success for human teams to be the most important: Does the automated system make them better at their work? Are team members performing at a higher level than they did previously? Can they apply their skills more creatively? Does the availability of automation technology allow teams to do things that they could not have done otherwise?
. . .
The General Motors vision for a factory of the future was productivity and flexibility without the need to light the way for workers. But what we have learned from companies at the frontiers of automation is that even if they could achieve something like lights-out, they would probably pass. They’ve learned that marrying productivity and flexibility requires humans to be in the loop, learning where technologies are working and where they can be improved. Companies are best served by a positive-sum automation that draws on the strengths of intelligent machines, managers, engineers, and line workers alike. The vision is not one without humans but one in which automated systems make humans more capable and more vital at work.