If you bought an AI tool three months ago and it's basically decoration now, you've probably asked yourself why isn't my team using the AI tool we bought. You watched the demo. It looked obviously useful. Then you handed it to your team and the usage numbers came back at near zero, and you're left wondering if your staff is lazy, scared of the technology, or just stubborn.
They're not. Salim Ismail argued this week that inside big companies, the people quietly slowing down AI initiatives aren't saboteurs or Luddites — they're doing their jobs exactly as designed. The security team that takes six months isn't being difficult, it's being measured on risk avoidance. The middle manager who buries the AI project isn't scared of change, he's optimizing for the metric his bonus actually depends on, and that metric has nothing to do with AI adoption. Ismail's point is that the whole system is behaving rationally. Nobody built a reason for these people to want the thing to succeed.
Your Software Is Fine. Your Incentives Aren't
Here's the part owners of small companies get wrong: you assume this is a big-company problem, something that happens with steering committees and enterprise procurement cycles. It's not. It's smaller and faster in a 15-person company, but it's the same mechanism. You don't have a committee slow-walking the rollout. You have one person — the office manager, the senior associate, the shop foreman — who has quietly decided the new tool isn't worth her time, and she's right, given what she's actually measured on.
This is not a training problem. You can run a two-hour workshop on how to use the tool and usage will still flatline a month later, because the workshop didn't change what happens to her paycheck, her performance review, or her Friday afternoon workload if she uses it or doesn't.
Why Your Office Manager Won't Switch Tools (And She's Right Not To)
Say you're running a dental clinic and you bought an AI scheduling assistant that's supposed to cut no-shows and fill gaps automatically. Six weeks later your office manager is still booking everything by hand in the old system. You think she's resisting change. Actually, she's been graded for eleven years on zero double-bookings and a calm front desk. The new tool moves faster than she trusts, and if it makes one mistake, that mistake is visible and hers. Nobody told her that adoption of the new tool counts toward anything she's evaluated on. So she does the rational thing: she avoids it.
Same pattern at a small hotel. You install an AI tool to handle guest messaging and upsells. The front desk staff keep answering everything manually. Why would they change? Their bonus structure rewards guest satisfaction scores, not response time, and they don't trust an AI reply not to say something wrong under their name.
The Senior Associate Who's Quietly Avoiding the New Drafting Tool
Law firms show this most clearly because billable hours make the incentive explicit instead of hidden. You buy an AI drafting tool that can turn a two-hour contract review into twenty minutes. Adoption should be instant. It isn't, because your senior associate bills by the hour and gets evaluated partly on hours logged. If the AI tool cuts her hours, and nobody has changed how she's evaluated, she is being asked to personally shrink her own numbers for the good of the firm. She won't do that, and honestly, you wouldn't either.
Manufacturers hit a version of this too. You put in an AI tool that flags maintenance issues before machines break. The floor supervisor doesn't act on the alerts because his metric is uptime this shift, not prevented failures next quarter that he can't take credit for.
Name Who Benefits Before You Roll Anything Out
The fix Ismail points toward for large organizations shrinks down cleanly for a 15-person company, because you have fewer people and shorter lines of sight. Before you buy or roll out anything, sit down and name, out loud, exactly whose current incentive is in conflict with this tool succeeding. Not in general — specifically, by name. The office manager rewarded for zero errors. The associate rewarded for hours billed. The supervisor rewarded for this shift's number, not next quarter's.
Once you've named it, you have two options: change what that person is measured on, or accept that the tool will sit unused no matter how good it is. There's no third option where good intentions or a compelling demo overcome a real incentive conflict.
What Actually Gets Measured Gets Used
The practical move is to change the scorecard before you change the software. If you want your senior associate to use the drafting tool, don't tell her to use it more — change how she's evaluated so time saved counts as a win, not a loss. Put a line in her review: contracts turned around faster, hours saved and reinvested in higher-value client work. If you want your office manager to trust the scheduling AI, give her explicit cover: tell her that a mistake caught and corrected within an hour is a success story, not a failure, and put that in writing somewhere she'll actually see it, like her quarterly check-in.
This costs you nothing but attention. You don't need a data science team or a change-management consultant. You need one conversation per key person, where you say plainly: here's what I'm now measuring instead, and here's why it benefits you specifically.
A 15-Person Version of an AI Adoption Plan
A workable version of this looks small on purpose. Pick the one tool you most want adopted. List the two or three people whose daily habits it has to change. For each one, write down, in one sentence, what they're currently rewarded for that conflicts with using the tool. Then change that one thing — the bonus line, the review question, the thing you praise them for in the team meeting — before you push the tool again.
Check usage weekly for the first month, not to police anyone, but to see whether the incentive change actually moved behavior. If it didn't, the incentive wasn't the real one. Real estate brokerages that have done this well didn't roll out AI lead-scoring to the whole team at once — they picked the one agent whose commission structure already rewarded faster follow-up, made her the visible early win, and let her results do the persuading that a company memo never could.
The tool was never really the hard part. The hard part is admitting that the people ignoring it are behaving exactly as you built them to.
