September 17, 2026, (Inside AI) — A large German bakery chain discovered that removing two daily operational checklists from half its stores led to a 2.7% increase in sales and a 20% drop in attrition among trained workers, according to a randomized controlled trial. The study, conducted by researchers including a professor of human resources, found no decline in quality and even improved customer reviews. Yet the same experiment revealed a hidden cost: for less experienced staff, removing the checklists caused a 20% rise in attrition. The findings, published in a working paper, challenge the assumption that more monitoring always improves performance and offer a nuanced view as AI makes surveillance cheaper and more pervasive.
The research comes at a time when companies face a critical choice. AI-powered tools can now track keystrokes, analyze video feeds, and monitor token usage in real time. The cost of control is plummeting, and the temptation to monitor everything is rising. But the bakery experiment shows that the benefits of monitoring are not universal. They depend on who is being watched and why.
“The math was simple,” said Linus Arauz, who operates 10 points of sale across two large airports and oversees nearly 100 food and non-food retail locations. “After going through disciplined control exercises, we found out the actual cost of time, overhead, and logistics invested in counting inventory across our stores was significantly higher than the loss itself.” Arauz decided against installing extensive monitoring systems, relying instead on a few cameras. His conclusion: control would cost more than it would save.
The bakery chain’s experience echoes that logic. Managers often focus on the direct benefits of monitoring, such as reduced theft and fewer errors. But the study highlights two types of costs. The first is direct: hardware, software, staff hours, and friction for honest customers. The second is subtler: how being watched changes behavior. Behavioral economists Armin Falk and Michael Kosfeld demonstrated in a classic paper that imposing control can signal distrust, causing employees to expend less effort. The bakery trial confirmed this in the real world.
Read: AI and Workplace Psychological Safety
When researchers removed two of the most-disliked checklists from half the stores, sales rose and experienced workers stayed longer. Customers praised the treated stores for faster service and better-looking goods. But newer employees struggled without the structure. The lesson is not that monitoring is bad. It is that monitoring must be matched to worker qualifications. A checklist that insults a master baker may be a lifeline to a novice.
“The real lesson is not that monitoring is bad. It's that monitoring must be matched to worker qualifications,” the researchers wrote. “The same checklist that insults a master baker might be a lifeline to a novice.”
This nuance is often lost as AI makes monitoring cheaper. At Amazon, Meta, and Uber, developers were encouraged to use more AI tokens, and their consumption was ranked on internal leaderboards. Amazon shut down its token leaderboard after employees inflated usage to climb it, driving up computing costs without producing more work. Meta suspended its Model Capability Initiative, which logged keystrokes, clicks, and screenshots to generate AI training data, after internal data turned out to be more widely accessible than intended. These examples show that cheap monitoring does not make the costs of control vanish. It only makes them easier to ignore.
To avoid what the researchers call “the control trap,” leaders should audit existing monitoring systems and ask what each piece costs in money and behavior, and what it actually buys. They should adapt controls to the people they touch. The bakery chain, for instance, kept one checklist after the experiment. Dutch teaching hospitals have since 2006 given residents written statements of mastery for defined clinical tasks, allowing them to work without supervision. Oversight falls away one task at a time, on terms everyone understands up front.
Read: Wipro AI Push Frees Capacity
“Falling cost isn't a reason to surveil — it's a reason to choose more carefully than ever,” the researchers argued. The firms that thrive in the age of cheap monitoring will be those that consider both sides of the ledger. Healthy businesses need controls, but controls are not a strategy. As AI offers cheap control solutions, managers may be tempted to monitor more. But the bakery experiment shows that in many cases, the problem is not a lack of control. It is a lack of leadership, training, or strategy. The most successful firms will be those that resist the urge to monitor simply because they can.
