By Andrew Miller
More than half of small businesses now use AI regularly, and most owners say it’s paying off.
87% of those users report a positive impact on the business, according to OnDeck and Ocrolus’s latest small business report. That’s the number ownership sees. It’s not the number the employees doing the day-to-day work would give you.
For some employees, checking AI’s work feels like another job entirely. Every output has to be reviewed, then either accepted or corrected. Do that often enough in a day and the checking itself starts to wear people out more than the original task would have.
Researchers at Boston Consulting Group and UC Riverside found this kind of oversight has a real cost, although not the one people expect. Burnout actually drops when AI handles repetitive, low-stakes work. It rises when employees have to exercise judgment on what the AI produced, and rises further once someone is juggling three or more AI tools at once. They named the result “brain fry.”
Fatigue changes how people check. Early in the day, most people read carefully. By the tenth or twentieth AI output, they’re skimming. They question less and accept more. Eventually the pattern flips entirely: instead of verifying and then trusting, people trust first and skip the verifying.
Researchers Steven Shaw and Gideon Nave at Wharton have a name for what happens next. They call it cognitive surrender, it’s the point where someone stops evaluating an AI’s answer and just adopts it as their own, without registering that a decision got handed off.
To test it, they had people solve reasoning problems with an AI that had been secretly programmed to give wrong answers part of the time.
On their own, people got about half the questions right. Paired with an AI that happened to be correct, they did significantly better. Paired with an AI that was wrong and trusted anyway, they did significantly worse — worse than the people who never had an AI to consult at all.
Trusting without verifying doesn’t just cancel out the benefit. It leaves people behind where they started.
Part of what makes the trust so easy to give is that AI never sounds unsure. A tired employee double-checking a colleague’s work can pick up on a shrug, a hedge, an “I think.” AI delivers a wrong answer with the same tone as a right one. There’s no tell to notice, especially once noticing takes more energy than a person has left.
The one thing that reliably brought the scrutiny back: giving people a reason to be accurate, paired with quick feedback on whether they got it right. People started catching bad AI answers again once there was something at stake and something to learn from immediately.
That’s the actual lever for a business owner handing AI tools to a team. Build in a reason to look twice and a way to know quickly when that second look mattered before fatigue sets in. Skip it, and the tool that was supposed to save time starts costing more than it saves.
AI isn’t an all-knowing entity. It can’t replace the real-world judgment an experienced employee brings to the table. Only when the two work together does either one become a real gamechanger.
P.S. A second look also matters for big decisions like where your business gets its next round of funding. Signet can help you find the right fit.Get in touch here.