Latest / The Productivity Podcast with Fexingo: Output, Efficiency, and Long-Term Economic Growth / The Productivity Trap in Routine Work Automation
Transcript
- Lucas: There's a finding from 2024 that still bothers me every time I think about it. Researchers at MIT looked at a mid-sized insurance firm that rolled out robotic process automation for its data-entry team — the kind of software that copies numbers from one spreadsheet to another, fills in forms, that whole category of mind-numbing work. Luna: And I'm guessing the results weren't the soaring productivity story management had in mind. Lucas: Exactly. Over six months, overall output actually dropped by six percent. Not just stayed flat — fell. And the automation itself worked perfectly. No bugs, no errors. The software was doing exactly what it was supposed to do. Luna: So the problem wasn't the tool. It was something about how the people reacted to having their boring work taken away. Lucas: That's the puzzle. And if walking through the economy with us has made something click — maybe it's a new way of thinking about your own work, or a curiosity about how these systems really behave — listener support is what keeps this show ad-free and independent. If that sounds worthwhile, you can find us at buy me a coffee dot com slash fexingo. No pressure, just a quiet way to say this matters to you. Luna: Yeah, it genuinely helps us keep digging into these stories without chasing sponsors. And now back to that insurance firm — what did the researchers find was going on underneath the numbers? Lucas: They interviewed the workers and tracked their activity logs. What they found was a kind of cognitive disengagement. When you automate the low-level decisions — which field to copy, which button to click — the human operating alongside the software stops paying attention. They go into a passive monitoring mode. And when something unexpected happens, they're slower to react than they were before, because they've lost the rhythm of the work. Luna: So the automation created a gap — the person was still there, but they weren't really present. Lucas: Right. The researchers called it the 'productivity valley' — a U-shaped curve where moderate automation boosts output, but full automation of routine tasks can actually harm it, because the human's attention collapses. The optimal point, they found, was around 40 to 50 percent task automation, where the software handles the repetitive drudgery but leaves enough cognitive demand to keep the person engaged. Luna: That feels counterintuitive. Most managers I know think 'automate everything you can' is the goal. Lucas: It's the dominant mindset. And it works fine when the output is purely mechanical — a factory robot that welds the same joint ten thousand times. But when you've got a knowledge worker who's supposed to handle exceptions, catch errors, make judgment calls, you can't afford to have them checked out. Luna: This actually reminds me of something from early industrial psychology. There were studies in the 1920s at a Western Electric plant — the Hawthorne experiments — where they changed the lighting, then changed it back, and productivity went up both times. The explanation was that workers responded to attention, not just the physical conditions. Lucas: That's a perfect parallel. The Hawthorne effect is exactly about the human element. But what the MIT study adds is that the human element isn't just about feeling watched — it's about having enough cognitive load to stay in the game. If you take away too much, the brain treats it like a boring lecture and starts to drift. Luna: So what does a company do? Say they have a team of data processors handling invoices. Do they deliberately leave some manual steps in place? Lucas: Some companies are doing exactly that. I came across a logistics firm in Denmark — they handle freight documentation for a bunch of European ports. They automated about 60 percent of the data entry, but they deliberately left the final verification step to a human, and they rotated which documents were verified manually. That way the workers always had to stay alert because they never knew which file they'd need to double-check. Luna: That's clever. It's like keeping the operator in the loop by design, not as an afterthought. Lucas: Exactly. And their productivity actually improved by 12 percent after the automation rollout — but they avoided the valley by preserving that cognitive engagement. Compare that to the insurance firm in the study, which automated 80 percent of the steps and left the workers with almost nothing to do except monitor. Luna: I wonder if this also explains why some companies that go all-in on automation see a spike in errors that they can't explain. Lucas: It probably does. Because the errors aren't coming from the software — they're coming from the human who missed the edge case because they'd been staring at a screen for three hours without having to think. The MIT study documented a 20 percent increase in exception-handling time after the full automation was introduced. Luna: Twenty percent. That's a huge drag on the supposed efficiency gain. Lucas: It wipes out the benefit entirely. The company spent money on the automation, saw output drop, and had to dedicate more senior staff to clean up the mess. In the end, they actually rolled back some of the automation to a 50 percent level, and productivity recovered. Luna: So the lesson is: automate, but don't automate to zero. Lucas: I think that's a great way to put it. The question every manager should ask before implementing any automation is: what will the human be doing with the remaining time? If the answer is 'staring at a screen waiting for something to go wrong', you've probably gone too far. If the answer is 'analyzing exceptions, making judgments, or doing higher-level work that requires attention', you're in the sweet spot. Luna: That reminds me of Toyota's approach to automation — they call it 'autonomation' or 'jidoka'. They intentionally design machines that stop when something goes wrong and require a human to diagnose the issue. It's not about removing the human; it's about augmenting their ability to spot problems. Lucas: Toyota's system is a brilliant counterexample. They've been doing this since the 1960s. Every worker on the line has a cord they can pull to stop the entire production line if they see a defect. That's the opposite of passive monitoring — it's active, engaged, and it works because the person feels responsible for the outcome. Luna: And Toyota's productivity numbers speak for themselves. They consistently have among the lowest defect rates and highest throughput per worker in the auto industry. Lucas: So the thread here is that productivity isn't just a matter of replacing human effort with machine effort. It's about designing a system where both complement each other. The MIT study gives us a clear, data-driven warning: if you strip away too much cognitive demand, you lose the human advantage. Luna: What about knowledge work that's less routine than data entry? Say, a financial analyst who runs reports — some of that could be automated, but the interpretation still requires a human. Lucas: Same principle applies. If you automate the report generation but leave the analyst with just a dashboard to glance at, they might stop noticing anomalies. I've seen that happen at a mid-size asset manager where they automated portfolio rebalancing — the system worked fine, but the human oversight became so passive that they missed a compliance violation for two weeks. Luna: So the fix would be to have the system flag exceptions but require the analyst to manually confirm each one, rather than auto-approving. Lucas: Exactly. Force a decision point. That's the 'human in the loop' principle that the Danish logistics firm used. It's not about maximum automation; it's about optimal automation. And optimal is usually less than maximum. Luna: I want to come back to that phrase 'productivity valley' — it's a useful mental model. If you're a manager listening, how do you know where on the curve your team sits? Lucas: The researchers suggested a simple diagnostic: measure exception-handling time before and after automation. If it goes up, you've probably entered the valley. Another signal is worker self-reports of boredom or disengagement — the study found that workers who reported high boredom were 30 percent slower at catching errors. Luna: So the boring work isn't just unpleasant — it's actually counterproductive. Lucas: It can be. And I think that's the core insight of this episode: the goal of productivity design shouldn't be to eliminate all boring work. It should be to eliminate the kind of boring work that makes people stop thinking. Keep a little bit of the good kind of routine that maintains engagement. Luna: That's a subtle but powerful distinction. And it challenges a lot of the hype around full automation. Lucas: It does. The next time you hear a vendor promise 'zero-touch processes', I'd ask: what happens to the people who used to touch those processes? If the answer is 'they'll supervise', you might want to test it carefully before rolling it out across the whole company. Luna: Or design it so they can't supervise passively — they have to touch some of it. Lucas: Exactly. That's the practical takeaway. Don't automate to zero engagement. Automate to the point where the human's attention is preserved — and then stop.