On October 20, 1984, The New York Times ran an article headlined, “GM Factory of the Future Will Run with Robots.” In it, Roger Smith, then GM’s CEO, claimed that automation would save the company from increasingly formidable Asian competitors.
But that didn’t happen. Smith’s robotic factories struggled to match the productivity of their human-run counterparts. Robots sometimes painted each other instead of cars or welded doors shut. And they carried much higher costs.
Today, the assembly of automobiles and countless other products is done primarily by robots. Smith had the right idea; he just went about it the wrong way. Artificial intelligence poses a similar challenge.
A recent report by our colleagues at MIT suggests that despite the $30 billion-$40 billion currently being invested in enterprise AI, 95% of pilots are getting zero return. Just as automation ultimately changed manufacturing, AI will undoubtedly reshape how companies operate; however, GM’s experience highlights the pitfalls of not thinking about its implementation carefully. Throwing technology at problems without understanding how work gets done day-to-day is a surefire way to waste money and breed cynicism.
Take a cue from Taiichi Ohno, the engineer known as the father of the Toyota Production System. He argued for “autonomation:” or automation with a human touch. Here’s how leaders can put his insight into practice with AI:
One of the students we taught at MIT Sloan School of Management likes to say, “There are few ways to lose money faster than automating a process you don’t understand.” That was Smith’s first error.
Automotive assembly plants are complex environments. Every process combines formal procedures and countless local refinements to get work done. Most of these tweaks, while necessary, are invisible to people one level up, let alone the CEO.
Knowledge work is even harder to map and is often shaped by thousands of micro adjustments. Consider all the emails and hallway conversations needed to move any decision forward. Leveraging automation requires understanding both the way work is supposed to be performed and how it’s actually done.
Successfully using AI requires a similar approach. You have to understand the work, otherwise you risk creating tools that, as the MIT report concluded about current AI applications, are “…brittle, overengineered, or misaligned with actual workflows.”
Smith’s second mistake was going too big, too fast—trying to replace entire systems overnight rather than proceeding incrementally with small, focused experiments.
Toyota pinpointed jobs where robots could make the work better by doing things like eliminating unsafe activities and physically taxing jobs. Then they ran experiments. Safety and productivity improved without upending the whole system, which allowed them to learn how to design work that robots could do repeatably. With this knowledge in hand, using robots for the next round of changes was easier and less disruptive.
The AI analogy is clear: repetitive tasks are dull and create the mental equivalent of repetitive stress injuries. Look for processes that are predictable and repeatable. Start where boredom is high and variability is low then use these simpler automation successes as learning experiences toward automating more sophisticated, complex work.
AI will never grasp the full context of your organization or the surrounding social and political dynamics. AI only knows what it has learned from experience. You still need employees who know the work and the organization to oversee AI to make sure its learning is headed in the right direction.
There’s little doubt that AI will eventually eliminate jobs, but if your company hopes to grow and thrive, choose this as a last resort. Smith didn’t think this way. His tenure was marked by plant closures and job losses. He famously told auto workers, “Every time you ask for another dollar in wages, a thousand more robots start looking practical.”
This is misguided. The “machines versus people” dynamic has fueled labor tensions, slowed technology adoption, and hurt organizational performance for over a century. It’s also bad business. Technology should improve productivity and fuel growth, not just slash costs.
AI frees up capacity. Use this newly available bandwidth to dust off ideas that have been sitting on the shelf: new services to offer, new markets to enter, and nagging problems to finally solve. Position employees where their skills are strongest; you know them, and they know the business.
Our approach requires a strong stomach, at least initially. At first, it’ll feel too small and too slow, especially when competitors boast about “doing AI everywhere.” But as you clear away work that is easily automated, building skills along the way, and delivering returns on the AI investment, more complex challenges will appear. Rinse and repeat with the next opportunity, ensuring that AI is not just cutting costs, it is helping you redesign and grow the business.
Much as robots are everywhere in factories now, AI will find a permanent place in most organizations. Your company will get there faster and with less heartache if you understand how work gets done, start with small experiments and prioritize growth over cuts.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
This story was originally featured on Fortune.com
