Buying an AI Tool
AI adoption in a small business is decided before you buy
Quick answer: AI tools in small firms rarely fail because they do not work. They fail because nobody opens them in week three. That is not a training problem or a change-management problem. It is a decision problem: the people who do the job were not in the room when the tool was chosen, so the tool was chosen for a job they do not recognise, in a place they do not work. Involving them in the choice, running the trial through them, and scoring adoption before the purchase fixes most of it. The free method for doing that, including the twelve adoption questions, is at practicalmethods.ca/guides/buying-an-ai-tool.
The failure that does not show on the invoice
An AI tool that fails loudly is rare. The common failure is quiet. The subscription renews, the seats stay assigned, and by the end of the first quarter one person uses it. Usually the person who bought it.
Firms respond to this with training sessions, lunch-and-learns and a Slack channel of prompt tips. Those treat the symptom. The tool went quiet because of decisions made before anyone logged in: which job it was for, where the output lands, and who was asked. None of those can be fixed with a training session. All of them can be fixed by making the purchase differently.
Who was in the room
The most reliable predictor of adoption is a question most firms never ask: who asked for this?
When the owner chooses a tool after a conference or a post, and rolls it out, the tool starts from behind. It is solving the owner's picture of the job, not the job. When the people who do the job weekly pick the tool, or at least run the trial and score it, they have already decided to use it before the invoice arrives. The purchase and the adoption are the same act.
This is why the method treats the decision as a group exercise rather than a procurement task. The playbook version of it is a ninety-minute session with the people who do the job in the room, and the reason for the room is not consensus. It is that adoption is being scored by the people whose adoption it is.
What "involving others" actually means
Involving people is not a survey and it is not a demo. It is three specific things, in order.
Eight lines, before any vendor page is opened: the one job, who does it weekly, how many people will really log in, the ceiling, the data it touches, the systems it must sit inside, what "worked" looks like at ninety days, and who signs. The people who do the job know lines one, two and six better than the owner does. If they write those lines, the tool that gets chosen will be for the job they actually do, in the place they actually do it.
Not the buyer. Give three people two weeks with each shortlisted tool: the person who does the job most, the person most sceptical of it, and one in between. Say nothing. No reminders, no check-ins. At the end of week three, count who used it without being asked. That number is the adoption score. No vendor onboarding session will change it.
The rating is one to five on twelve questions, from the trial. The people who ran the trial hold the pen. The owner can disagree with a score, in the room, with the evidence on the table. That argument is worth more than any rollout plan, because it happens before the money is spent.
The twelve questions, briefly
The full list is in the adoption article. The six that decide most purchases:
| Question | What it predicts |
|---|---|
| Who on the team asked for this? | Whether adoption starts ahead or behind |
| Did the people who will use it daily run the trial? | Whether the score is real |
| How many used it in week three, unprompted? | The adoption score itself |
| Is it inside the tool they already have open, or a new tab? | Whether it survives a busy afternoon |
| Who owns it after purchase? Name a person. | Whether anyone reviews seats or runs the 90-day check |
| What is the plan for the person who refuses to use it? | Whether the tool has a fit problem nobody has named |
The last one is the question nobody wants to write down. There is always one person who will not use it. Ask what would have to be true for them to. The answer is nearly always specific and nearly always a real weakness in the tool, not in the person. Write it down. It is a finding.
Why the bundled tool keeps winning this criterion
The assistant already inside Microsoft 365 or Google Workspace scores well on adoption without trying. Same tab. Minutes to first output. Nothing to set up. Nobody has to be told where it is.
Some buyers feel that is unfair to the standalone tools. It is not. It is the reason standalone tools go quiet in small firms, and it is the reason the method starts by running the job through the tool you already pay for. A standalone tool that beats the bundled one on the job by a wide margin can afford to lose on adoption. One that beats it narrowly usually cannot, and the people in the room will tell you which it is.
What this looks like when it works
A twenty-two person engineering firm wanted an AI tool for drafting proposals. Everyone assumed the answer was a general assistant seat. Six people wrote the shape, three ran the trial on three tools, and the scores came in four points apart. The general assistant wrote the best draft. The bundled assistant was the one five of six people were still using in week three, because it drafted inside the document the proposal ships from.
They kept the bundled tool, built one template, and spent nothing. The decision held because the people who would have had to adopt the other tool were the ones who scored it. Nobody had to be trained into anything.
The full method is five free articles, at practicalmethods.ca/guides/buying-an-ai-tool. Article three is the adoption criterion, with all twelve questions.
Frequently asked questions
Why do employees stop using AI tools?
Usually because the tool lives in a new tab, needs a prompt expert to get value, or produces output that has to be copied somewhere else. These are decided by which tool was bought and for what job, not by how it was rolled out.
How do I get my team to adopt an AI tool?
Involve them before you buy. Have the people who do the job write the shape, run the trial, and score adoption from the trial. A team that chose the tool has already adopted it.
How long should an AI tool trial be?
Two weeks of real use, with three people, and no reminders. Count who is still using it in week three. That is the adoption score.
Should the owner or the team run the trial?
The team. A trial run by the buyer produces a score from the buyer, and the buyer does not do the job.
What if only one person uses the tool?
Then it is a one-seat tool. That is a fine outcome if it is priced as one seat. The mistake is buying twenty seats on the strength of one person's enthusiasm.
Does training improve AI adoption?
Sometimes, briefly. Training fixes not knowing how. It does not fix a tool that lives in the wrong place or produces output that has to be moved. Score the structural questions before buying instead.
How do I handle an employee who refuses to use the AI tool?
Ask what would have to be true for them to use it. Treat the answer as a finding about the tool, not a complaint about the person. It is usually a fit or path problem the gates can catch.
Is the AI already in Microsoft 365 or Google Workspace better for adoption?
Usually. It is already open, needs no setup, and its output lands where the work is. Run the real job through it before buying anything standalone, and put it on the scorecard as a real option.
Criteria: Buying an AI Tool
Five articles. Narrowing, the job, the adoption, the trust, the deal.
Read the free methodThe list is not the decision. It is how you find the one or two things that are.
Not a review, not rankings, not software advice. A method.
