There is a specific pattern to unused AI licences. The business buys them, sends an announcement, runs a lunchtime demonstration of impressive capabilities, and usage peaks in week two and decays from there.
The cause is not resistance. It is that people were shown a tool and never given a job. Here is a format that reliably fixes it, and it takes half a day.
Before the session: gather real work
Ask everyone attending to bring one task they do weekly that involves writing, reading or extracting something. Not a hypothetical. The actual thing, with the actual documents, that they will do again next Tuesday.
This single instruction is the difference between training that sticks and a nice presentation. Every exercise below uses their own work.
Group people by the kind of work rather than by department. The person in finance chasing invoices and the person in operations chasing subcontractors are doing the same job and should be in the same room.
Part one: what it is actually good at (30 minutes)
Do not start with capabilities. Start with the boundary, because the fastest way to lose a room is to over-promise and have somebody find the limit in week one.
Cover the four things that genuinely work — drafting, summarising, extracting structure, answering from your own documents — and then be equally specific about what does not. It does not know today's prices. It does not know what your company decided last Thursday. It cannot do arithmetic reliably on a long column of figures. It will agree with you if you ask a leading question.
People trust a tool more, not less, when somebody has told them where it stops.
Part two: context beats cleverness (30 minutes)
Skip prompt engineering. Modern models handle plain instructions well and the elaborate techniques circulating in 2023 are mostly obsolete.
Teach one idea instead: the output is only as good as the context you supplied. Four things to include, every time:
- Who it is for. A quote for a facilities manager reads differently from one for a finance director.
- What good looks like. Paste in a previous example of the thing done well.
- The constraints. Length, tone, what must be included, what must not be mentioned.
- The source material. The actual documents, not a description of them.
Demonstrate this once with a deliberately bad prompt and then the same request with context added. The difference does the teaching.
Part three: hands on their own task (90 minutes)
This is the session. Everything before it is preamble.
Everyone does their own real task, with the trainer walking the room. They will get stuck in ways no prepared exercise would have produced, and those are the moments people remember.
Two rules for whoever is running it. First, do not take the keyboard. Second, when somebody produces something disappointing, ask what context was missing rather than supplying a better prompt yourself. The goal is that they can do it again on Wednesday without you.
Everyone should leave the room able to name the specific task they will do differently next week. If they cannot, the licence will be dormant by month two.
Part four: checking and the rules (30 minutes)
Finish with the two things that keep the business safe, taught as habits rather than as policy.
Checking. Every number, date, name and reference gets verified against a source. Ask which document a grounded answer came from. Read anything customer-facing aloud before sending. Ten minutes, framed as professional craft rather than compliance.
What must never go in. The green, amber and red categories from your AI policy, with two or three worked examples from your own business. People remember the examples and forget the categories, so make the examples good ones.
Six weeks later: the session that matters most
Book it on the day of the first session, because it never gets booked afterwards.
An hour, same group. Everyone brings one thing that worked and one thing that did not. By this point the questions are real and specific, people have hit genuine limits, and the room teaches itself more than any trainer does.
This follow-up is what separates businesses where AI became normal from businesses that ran some AI training once.
What to measure
Not licence logins. Ask three questions at the eight-week mark:
- Name the task you now do differently. Can they?
- Roughly how long did it take before, and how long now?
- What have you had to stop using it for?
The third question is the honest one. A team that can answer it has genuinely learned the tool. A team that says it works for everything has not been paying attention.
If you would like this run for your team, using your systems and your actual work, our IT consultancy team delivers exactly this format for UK businesses. Get in touch and we will tailor the exercises to the work your people actually do.
