Why licences alone do not work
Many businesses buy AI licences, send a launch email and find a month later that most people tried the tool once and gave up. Usually the first attempt went poorly: a vague question produced a generic answer, and the person concluded AI was not useful for their job. Others overuse it, accepting whatever it produces without checking. Both problems come from the same gap. Staff were given a powerful tool with no guidance on how to use it for their work. A small investment in practical training makes the difference between an expensive licence and a tool that saves real time.
Start from real work
Before the session, ask each person or team for two tasks they would like help with: replying to common customer questions, writing site reports, summarising supplier contracts, preparing board papers. Build the training around those examples. A receptionist and an estimator need different demonstrations, and both learn more from their own work than from generic examples about planning a holiday. Small groups work better than one big presentation. Thirty to forty-five minutes with four or five people from similar roles, each trying the tool on their own tasks with someone helping, gives far better results than a long lecture.
Teach how to ask
The single biggest improvement comes from better instructions. Teach a simple pattern: say who you are and who the output is for, give the relevant background, describe exactly what you want, and specify the format and length. Compare write a reply to this customer with: I run the service desk at an electrical wholesaler; draft a short, polite reply to this customer whose order is late, apologise, give the new date of Thursday, and offer a phone call if they need it sooner.
Show people how to iterate. The first answer is rarely the best. Asking the tool to make it shorter, more formal, more specific or to try a different approach is normal, and often quicker than starting again. Show them how to give the tool useful material to work from: paste the customer's actual email, the relevant clause, the bullet-point notes from site. AI writes far better when it has specific content rather than guessing. At the same time, remind them that pasted material must not include personal or confidential details, linking the prompting skill directly to the information rules.
Teach how to check
Make checking part of the habit, not an afterthought. Show real examples of AI getting things wrong, ideally from your own industry: an invented regulation, a calculation error, a summary that missed the key exception. Then teach a simple routine: check names, numbers, dates and claims against the source; open any references it gives; ask yourself whether the output says anything you did not provide or already know to be true.
Explain that responsibility does not change. If an email goes out with an error, it does not matter that AI wrote the first draft. That message lands better when managers model it themselves by checking openly. Give people permission to stop. If the tool is not helping with a task after two or three attempts, doing it the usual way is fine. Some work is quicker by hand, and part of using AI well is recognising that early rather than spending twenty minutes coaxing a mediocre draft out of it.
Cover the rules and the risks
Walk through your AI use policy with concrete examples: what information is fine to enter, what must never be entered, which tools are approved and how to anonymise a request. Show where AI is the wrong tool, such as final legal advice, decisions about people or anything where the facts must be current and exact. Include a section on AI-enabled scams. Staff who use AI themselves understand quickly how convincing fake emails and voices can be, which reinforces the rule that payment changes are always checked by phone on a known number.
Keep it going after the session
Create a shared space, such as a Teams channel or a page on your intranet, where people post prompts that worked and mistakes they caught. Nominate one or two enthusiastic staff as go-to people for questions. Check in after a month: who is using it, for what, and where are they stuck?
Run a short refresher every few months to cover new features and share the best examples from the team. New starters should get the same basic training in their first weeks, as part of onboarding, so habits stay consistent as the team changes. Measuring a few specific outcomes, such as time spent on monthly reports before and after, helps decide where to extend training next.
In short
- Licences without training lead to either abandonment or blind trust.
- Train small groups on their own real tasks, not generic examples.
- Teach a clear prompting pattern and the habit of iterating.
- Make checking routine and keep sharing examples after the session.