There is a familiar shape to AI training in a small business. Somebody books a half-day. The team is shown a series of impressive demonstrations. Everyone leaves interested. Three weeks later, two people are using it daily, the rest have gone back to exactly how they worked before, and the person who paid for the training concludes their team is not ready.
The team was ready. The training was fine. What was missing was everything that happens afterwards, and it is not a large amount of work. It is just different work from the kind that gets booked in a diary.
The short version
- Teach one task, not a tool. "Here is ChatGPT" changes nothing; "here is how we do meeting notes now" changes something.
- Two people decide whether it sticks: the enthusiast and the sceptic. You need both, for different reasons.
- Budget six weeks of follow-up, not a training day.
- Make the new way the official way, or the old way wins by default.
Why the half-day does not work
Three reasons, and they compound.
It teaches the tool, not the job. Being shown what an AI assistant can do in general leaves everyone with a vague sense of possibility and nothing to do on Monday morning. Possibility does not survive contact with a busy week.
The new way is slower at first. Always. The first three times you draft a customer reply with AI it takes longer than typing it, because you are learning what to tell it. Anyone under time pressure will abandon a method that is currently slower, and they are being sensible. Unless somebody has told them to expect that fortnight, they will conclude it does not work.
Nothing changed except the training. The process is the same. The template is the same. The expectations are the same. So the old way remains entirely available, and it is faster today. Habits do not lose to better ideas, they lose to better-supported ideas.
Start with one task
Pick a single recurring task and change how the business does that one thing. Everything else follows later.
Good first candidates share four features: it happens weekly or more, more than one person does it, everyone finds it dull, and a mistake is cheap. On that test the usual winners are meeting notes, first drafts of routine customer replies, and turning rough notes into a tidy document.
Bad first candidates: anything a client pays for directly, anything where an error is expensive, and anything only one person does. You want the first attempt to be low stakes and widely visible.
There is a full worked example of one such task in AI meeting notes, which is the most common place to start for a reason.
The two people who decide the outcome
Every small team has both, and the mistake is to focus entirely on the first.
The enthusiast has already been using it at home for months. They will adopt anything you suggest, immediately, and their example will not persuade anybody, because everyone else has already filed them under "they like that sort of thing". Their real value is as your builder: let them work out the prompt, the format and the setup, and write it down for everyone else. Give them that job explicitly.
The sceptic is the one who matters. Often your most experienced person, often the one whose standards everyone quietly uses as the benchmark. If they use it, it is real. If they refuse, it is a fad the boss is into.
Do not try to convert them with enthusiasm. Ask them to test it and report back honestly, including everything wrong with it. Two things follow. They engage properly, because you have asked for their judgement rather than their compliance. And their list of faults is the most useful document you will get, because it is the same list everyone else is thinking privately.
A six-week shape that works
| When | What happens |
|---|---|
| Before | Approved tools bought on business tiers, one-page policy written, one task chosen |
| Week 1 | Ninety minutes together on that one task only, using real work from the business. Everyone does it themselves; nobody watches a demonstration. |
| Week 2 | Enthusiast writes up the agreed prompt and format on one page. It goes where the team already looks. |
| Weeks 2-4 | Fifteen minutes at the existing team meeting: what worked, what was annoying, what it got wrong. Not a progress report. |
| Week 4 | Sceptic reports back honestly. Fix what they found, publicly. |
| Week 6 | Decide: this is now how we do it, or we tried it and it did not suit us. Say which, out loud. |
| Then | Only now pick a second task. |
Week six matters more than it looks. An initiative that is never formally adopted and never formally dropped simply fades, and the next thing you introduce is met with the memory of the last one that quietly stopped.
What to teach in the ninety minutes
Not a tour of features. Four things, all on real work from your own business:
- Context before instruction. The single biggest determinant of output quality, and the thing everyone gets wrong. Show the same request with and without a paragraph of background. The difference does the teaching for you.
- How it fails. Deliberately get it to invent something in front of everyone. People trust a tool more, not less, once they have seen its failure mode with their own eyes. There is more on this in AI hallucinations.
- The redaction habit. Show what a request looks like with the customer's details taken out. Ten seconds, demonstrated once, becomes automatic.
- The one task, done for real. Everyone does it, on their own screen, on their own actual work. Not a worked example, their own.
Ninety minutes is enough for those four. A full day is worse, not better, because it adds material nobody will use and dilutes the four things that matter.
The three sentences that make it stick
These come from you, not from a trainer, and they need saying plainly.
"It will be slower for the first fortnight. That is expected, keep going." Without this, everyone who experiences the slow fortnight concludes they are doing it wrong.
"This is how we do meeting notes now." Not "feel free to try". An optional new process competes with an established one and loses.
"Nobody here is being replaced by this." If that is true, say it, unprompted, early. Half the quiet resistance in any team is this worry, and it never comes up in a meeting. If it is not entirely true, say what is true instead. People can work with honesty and cannot work with silence.
What not to do
Do not measure adoption by logins. Do not run a competition for the most creative use. Do not roll out five tools at once. Do not ask people to report their time savings, which produces flattering guesses rather than data. And do not let it become one person's hobby: if only the enthusiast uses it, you have not changed the business, you have given someone a nice toy.
Where to start if you are the one who needs convincing
Perfectly reasonable position, and quite common among owners who have watched three previous technologies get oversold. The cheapest way to form a view is to watch someone competent do real work with it and see whether the output holds up.
That is what the free AI Breakfast Club webinar is for. One hour, online, every other Friday morning, then an open floor for questions. Gary builds live, on screen, and leaves the mistakes in, because the mistakes are the part you need to see before you ask a team to trust it.