AI adoption is the only AI metric that pays the bills. An AI system nobody uses is expensive noise.
The renewal notice arrives in October. Fifty seats of an AI platform, billed annually, auto-renewing in thirty days. Someone thinks to pull the usage report before signing, and it shows nine people logged in last month. Four of them logged in once.
At the same company, in the same month, the estimator is drafting scope letters with a free AI app on his phone. The office manager has a chatbot open in a browser tab all day. A project lead uses one to summarize the inspection notes she used to type up at night. None of them are using the platform the company paid for. All of them are using AI.
That contradiction is the most useful thing in the building. It is also the thing most owners misread.
The explanation everyone reaches for
When an AI rollout stalls, the explanation usually arrives before anyone asks a question. People resist change. The team is set in its ways. Some of them are afraid the tool will replace them. Send a reminder, schedule another training session, and maybe tie usage to performance reviews.
There is some truth in it. Change is uncomfortable, and nobody enjoys learning a new system while doing a full day’s work. But the explanation has a problem it cannot get around: the same people supposedly resisting AI are using it every day. They simply are not using the one the company bought.
People who resist change do not go find a substitute on their own time and pay for it out of their own pocket. People who have a better option do. That is not resistance. It is a verdict.
What is actually happening
The research has a name for this now: shadow AI. When MIT’s Project NANDA studied AI in business last year, it found that only 40% of companies had purchased an official AI subscription, while employees at more than 90% of companies were regularly using personal AI tools for work. The same study found that the company-sanctioned projects were the ones stalling.
The pattern has only grown since. In June, PagerDuty published a survey of 1,250 office professionals across four countries and found that two-thirds had used AI tools at work that they believed their company did not permit. Nearly three in four said they understood AI better than their own technology teams.
There are real risks in that, and I will come back to them. But look past the policy problem for a moment and the signal is clear. Employees are not avoiding AI. They are choosing it, deliberately, every day, for specific tasks. They are telling the company exactly where AI fits their work and which tools fit the way they actually do it. Most companies are treating that information as a compliance violation rather than the most honest requirements document they will ever receive.
Why capability is the wrong measure
AI tools are sold on capability. The feature list, the integrations, the benchmark scores, the demo where it handles forty kinds of tasks without breaking a sweat. Capability is easy to compare across vendors and easy to justify in a budget meeting, which is exactly why it dominates the buying decision.
But capability is potential. It describes what a tool could do if someone used it. The business only collects on the part that gets used, by real people, on real work, often enough to matter. A platform that can do forty things and is trusted to do none of them has produced exactly nothing, no matter what the feature list says.
The free app on the estimator’s phone has a fraction of the enterprise platform’s capability. It wins anyway, for reasons that never appear on a spec sheet. It is already open. It takes one step instead of six. It does not ask him to log into a separate system, find the right project, and reformat his notes before it will help. It fits the shape of his day.
That is the quiet lesson inside every abandoned AI rollout: adoption is not something that happens after the right tool is chosen. Adoption is how you know the right tool was chosen.
How Lean AI® builds adoption into the work
One of the five principles of the Lean AI® framework is that adoption matters more than capability. In practice, that means adoption is not a training problem to solve after launch. It is a design requirement from the first conversation.
Start where the work already is. Before evaluating any platform, find out who is already using AI and for what. Those employees have run the pilot for you, at no cost, under real conditions. Their habits show which tasks AI genuinely helps with and which ones it does not, more reliably than any vendor demo.
Fit the tool to the workflow, not the reverse. Count the steps between a person’s actual task and the moment the AI helps. If the official tool adds steps the free one does not, the free one will win every time. The fix is usually not a better tool. It is a shorter path to the one you have.
Bring shadow use into the light without punishing it. The risks are real. Customer data, pricing, and contract terms should not be pasted into personal accounts, and a business should know where its information goes. But banning tools people find useful mostly drives the use out of sight. A better approach is a short, plain policy about what data can go where, followed by approving and paying for the tools employees have already proven useful.
Train on the job, not the software. People do not need a tour of every feature. They need to see how AI handles the three tasks they do every day. Training built around real work gets used. Training built around the product gets forgotten by Friday.
Give every tool an owner and a number. Someone should be accountable for whether each AI tool is earning its keep, and that judgment should rest on the work it changed, not the licenses it holds. When the number says a tool is not being used, the honest response is to fix the fit or cancel the seats.
A five-question adoption check
Any owner can run this in an afternoon, without a consultant and without buying anything.
- What are we paying for, and how many people used it last month? Pull the actual usage report for every AI subscription. The gap between seats and active users is the first number that matters.
- What are people using instead? Ask directly, and make it safe to answer honestly. The goal is information, not enforcement.
- Which tasks keep showing up? Look for the same three or four jobs across different people. Those are the places AI already fits.
- How many steps does the official tool add? Walk through one of those tasks in the company’s tool and in the one people actually use. Count the difference.
- Who owns the result? For each tool worth keeping, name the person accountable for whether it changes the work. For each one nobody will own, plan the cancellation.
An owner who does this will often find two things at once: money being spent on AI nobody uses, and value being created by AI nobody approved. Both are worth knowing, and the second is usually worth more.
Back to the renewal notice
The fifty-seat renewal is still sitting there, auto-renewing in thirty days. The easy decision is to sign it and send another reminder about using the platform. The cheaper decision is to cancel it and pretend the question is settled.
The better decision starts with the estimator’s phone. Ask him what he uses it for, how many steps it saves, and what he would need for the company’s version to be worth opening instead. His answer will say more about the business’s AI strategy than the platform’s feature list ever could, and it will not cost fifty seats to hear it.
Ryan Gartrell is the creator of the Lean AI® framework and founder of Ryan Gartrell, P.A., an operational excellence advisory firm serving owner-led businesses. If you are paying for AI your team is not using, start with a conversation or take the free Operational Readiness Assessment.
