The pattern repeats across organizations of every size. A platform is chosen, licences are issued to a few hundred people, there is a launch session, and usage climbs for a fortnight. Then it settles at a number that embarrasses whoever signed the invoice — a committed minority using it constantly, and a long tail who logged in once.
The instinct is to blame the tool or the people. It is usually neither. It is that using AI well is a skill, skills are built by practice, and nothing in the rollout asked anyone to practise.
Why the launch workshop fails
A ninety-minute session can demonstrate a capability. It cannot build a habit. Attendees leave able to recognise what the tool does, which is not the same as knowing which of their own tasks to bring to it — and that translation step is where most people stall.
There is also a timing problem. Training lands before the need. Someone learns about document retrieval in March and first has a use for it in June, by which point the session is a vague memory and the interface has moved on.
People do not adopt a general capability. They adopt a specific thing that made Tuesday easier.
What changes the curve
- Role-specific worked examples. Not "how to write a prompt" but the actual task the finance team does every month, done in the tool, start to finish.
- Short and repeated. Ten minutes of practice several times a week beats a half day, because the retrieval is spaced and the habit has somewhere to attach.
- Assessment that means something. A completion tick records attendance. A timed exercise with a pass mark records capability, and people revise for it.
- A visible next step. Someone who has just finished should be able to see what comes after, or the sequence ends at the first module.
Where the training should live
Almost every organization does this in a separate learning system, and almost every organization pays for it twice: once in licence and once in the gap between where people learn and where they work. Two logins, two permission models, two sets of reporting, and a click between practising a technique and using it.
Putting the curriculum inside the workspace removes the gap. The example a learner just worked through is one tab from the document they need it for. Assignment follows the departments that already exist. Completion and capability show up in the same admin view as usage and spend, which is the view the person who approved the budget actually looks at.
The three groups, and why one plan fails all of them
Any group of five hundred people contains three populations with incompatible needs, and a single curriculum aimed at the middle misses all three.
- The already-fluent, perhaps one in ten. They do not need training; they need permission, model access and somewhere to publish what they work out. Make them the source of the worked examples rather than an audience for them.
- The willing majority. They will use it if someone shows them the version of their own job that it improves. This is where curriculum earns its money, and where generic material fails hardest.
- The unconvinced. Some have a considered objection worth hearing. Pushing training at this group early converts scepticism into resistance; leaving them alone until the middle group has visible results converts a few of them for free.
The common mistake is to design for the first group because they are the ones in the room when the decision is made.
Certificates, and whether they are worth anything
A certificate is worth exactly as much as the assessment behind it. A completion badge from a video nobody could fail is a participation trophy, and everyone involved knows it, which is why they sit unmentioned on internal profiles.
A timed exam with a pass mark, limited attempts and a serial number that a third party can verify is a different artefact. It is worth something to the individual, which is what makes them revise, and it is worth something to whoever has to demonstrate capability to a board or a client. If you are going to run assessments at all, make them the kind people can fail.
If nobody has ever failed it, it is not an assessment. It is attendance with better graphics.
What to measure
Licence utilisation is a weak proxy — it counts logins, not competence. Two better numbers: how many people completed a real assessment, and how many are still using the platform sixty days after they did. The first tells you the training worked. The second tells you it mattered.
If the first number is high and the second is low, the curriculum taught the tool rather than the job. That is a fixable problem, and a much more useful thing to learn in month three than in year two.