LibraryOperator answers8 min read
Why Nobody On Your Team Uses The AI You Bought
You bought forty seats and three people log in. That is not a training problem or a willpower problem, and another lunch-and-learn will not fix it.

You bought the whole team a seat. Forty of them, because the rep said the per-seat price drops at forty and everyone may as well have it. Two months later you open the admin dashboard out of plain curiosity, and there it is: three people logged in last month, and one of them is you. The other thirty-seven seats renew in October. At roughly thirty dollars a seat a month that is more than a thousand dollars a month buying nothing, and the part that stings is you cannot even be angry at the team. You know exactly why they are not using it. Nobody asked them to stop doing the thing that already works.
This is the most common question I get from operators who are past the "is AI real" stage and into the "I paid for it and it is sitting there" stage. An ops manager at a ninety-person distributor put it to me almost word for word: she has a chat window open all day, it saves her an hour easy, and she cannot get one other person on the floor to touch the thing. So here is the answer, and I am putting it first because you are busy and you have already read enough change-management posts.
The answer: you gave them a destination, not a removed step
Your team is not ignoring the tool because they need more training. Not because they are afraid of change. Not because they are quietly luddites who will retire before they learn a new app. They are ignoring it because you handed them a place to go, and the work was already in front of them.
People adopt a thing that removes a step from the task they are doing right now. They abandon a thing that adds a stop they have to remember to visit. That is the whole mechanism. A chat window in a separate tab is a destination. The official rollout, the one with your company logo on the login page, is also a destination. Both of them require a person to stop what they are doing, go somewhere else, figure out how to phrase the ask, wait, and then carry the answer back to the actual work. That is not a saving. That is a detour. The only reason you personally tolerate the detour is that you already believe in the payoff on the other side of it, so you pay the tax gladly. Your quoting clerk does not believe in the payoff yet, so to him the detour is just a detour, and he is correct.
That is why the fix is almost never another session. It is moving the AI out of the destination and into the step. When the follow-up email drafts itself inside the CRM record the rep already has open, adoption is not a battle, because there is nothing to adopt. There is no new place to go. The work got shorter. Nobody has ever needed a training session to make their afternoon shorter.
Why this is the answer and not just a nice theory
MIT's NANDA group spent the first half of 2025 looking at why enterprise AI keeps failing to land, across three hundred public deployments, fifty-two structured interviews, and a survey of senior leaders. Their headline number got quoted everywhere: ninety-five percent of enterprise generative AI pilots produced no measurable return. The number is not the interesting part. The reason is. They found the divide between the five percent that worked and the ninety-five percent that did not was not model quality, not the vendor, not the size of the budget. It was what they called the learning gap: the winners re-architected a workflow around the tool, and the losers bolted the tool onto the side of an unchanged workflow and waited for people to walk over to it.
Read that against your dashboard and it stops being an enterprise problem and starts being your problem, at your scale, exactly. You bolted the tool onto the side. Thirty-seven people looked at the bolt and kept working the way they already worked, because their way did not have a detour in it.
The same report has a second finding that should end the "my team just resists new things" story for good. While only about four in ten companies had bought an official AI subscription, employees at more than ninety percent of the companies surveyed were already using personal AI tools for real work. MIT called it the shadow AI economy. Sit with what that means. Your people are not refusing AI. Plenty of them are already pasting a customer email into a personal chat window on their phone at lunch, because in that one moment the tool removed a step they cared about. They will do that on their own, unpaid, unprompted, while your forty official seats sit dark. It was never resistance. It was placement. The personal tool met them inside a task that mattered to them. Your rollout asked them to come to it.
The honest take: where this stops working, and what it actually costs
This is the part the vendor will not tell you, so here it is.
First, embedding the tool into the workflow is real work, and it is precisely the work you were trying to skip when you just bought seats for everyone. Buying seats is easy, which is why it is the default and why the failure rate is what it is. Wiring the draft into the CRM, or the extraction into the accounting flow, or the summary into the ticket the rep already has open, takes integration effort, and sometimes a little building. That is the tax. You either pay it up front in setup or you pay it forever in a thousand-dollar-a-month seat bill that buys three logins. There is no version where the tax does not exist. The five percent paid it. That is the entire difference.
Second, some resistance is correct, and you need to be able to tell the two apart. When your best estimator will not let the tool near his quotes, the lazy read is that he is protecting his turf. Sometimes he is. But sometimes he is protecting a quality bar the tool genuinely cannot hold yet, and he is the only one in the building who can feel exactly where it fails. If you mandate the tool over that person's objection, you do not get adoption. You get compliance theater. He pastes the AI draft into the field, spends ten minutes fixing the two things it got wrong, and now the task takes longer than it did before, and he is right to resent it. A mandate cannot tell the difference between someone dodging work and someone holding a line. Placement can, because if the tool actually removed a step for him he would not need the mandate.
Third, adoption measured by logins is a vanity metric and it will lie to you in both directions. The shadow AI finding is proof the real usage is invisible on your dashboard. And a login is not a save. I would rather have three people whose Tuesday got genuinely shorter than thirty who dutifully open the app once a week so the number looks healthy in the board deck. Count tasks that got faster, or errors that stopped happening, or the after-hours calls that used to go to voicemail and now get answered. If you cannot point to a step that disappeared, nobody adopted anything, no matter what the seat report says.
Fourth, not everything on the team should be adopted by the team at all. Some jobs want one owner, not forty dabblers. The software regret data is brutal on this: most SMBs who regret a purchase blame the rollout, not the product, and buying everyone a seat is the rollout that fails most reliably. Spreading a tool across a whole team so it feels fair is how you guarantee it lands nowhere in particular. Sometimes the honest move is to put the capability behind one person who runs it as their thing, wire it into their workflow properly, and let the forty seats stay unbought.
None of this is a reason to stop. It is a reason to stop buying seats as the plan and calling the buying the rollout. The plan is finding the one step, in one workflow, that the tool can actually shorten, and putting it there so quietly that nobody has to be convinced of anything.
What to do about it
Go back to that dashboard, pick the one person out of your three who is genuinely faster now, and figure out what step the tool removed for them and where in their day it sits. That is your template. Not because it will scale to forty by Friday, but because it is the only pattern in the building that already works, and adoption spreads from a step that got shorter, never from a seat that got assigned. The question was never how to get people to use the tool. It was whether you bought them a tool or built them a detour.
Sources
Every claim above traces back to one of these. Go read them yourself.
- 01The GenAI Divide: State of AI in Business 2025
MIT NANDA / nanda.media.mit.edu / retrieved Jul 28, 2026
- 022025 Tech Trends: SMB software purchase regret and adoption
Capterra / capterra.com
Related reading
Nearest neighbours by meaning, drawn from the whole library rather than from matching tags. Some of these are from a different series on purpose.
The daily build
I built a place to catch the thing only you know before it becomes a problem
A metal fab shop owner went off grid for five days to see if his business could survive without him. It didn't make it past Tuesday. So today's build isn't a knowledge base, it's a ten-second habit for catching the thing only one person knows, before it turns into an outage.
Operator answers
Your Software Just Added AI. Do You Pay For It?
Three renewals this quarter, three new AI lines, priced anywhere from twenty nine dollars flat to a hundred and twenty five per seat. Which ones to pay for has almost nothing to do with how good the AI is.
Operator answers
What Actually Decides Whether ChatGPT Recommends You
Two of last month's calls said they found you by asking an assistant. The settings that decide whether your name can appear at all are free and take an afternoon, and the thing being sold to you as the fix is documented by Google as not being a requirement.