Scope the real task
Identify the sponsor, the participants and the problem. Describe the current workflow before introducing AI. Choose one task with a clear input, a useful output and a person who can judge the result.
Output: a brief with scope, examples and review criteria.
Prepare the conditions
Check accounts, permissions and data rules. Select fictional, anonymised or authorised examples. Agree how the group will practise, what tools are used and how to handle a missing access or a participant with less experience.
Output: prepared exercises and a workable environment.
Practise and challenge
Alternate short demonstrations with participants doing the work. Compare the results against the sources. Keep failed attempts visible: they show where instructions need improving and where human review remains necessary.
Output: a tested method, with its limits recorded.
Decide what happens next
Give the next experiment an owner. Record what needs to be checked, what time must be protected and when to review the trial. If follow-up is included, define its format and measures in the proposal.
Output: an owned next step, not an unassigned action list.
How do we know it was useful?
Evaluate a task before claiming a gain. Keep the starting point, the conditions and the review effort visible. Self-reported confidence and measured time are different signals.
| Question | A useful measure | What it does not prove |
|---|---|---|
| Was it reused? | A completed repeat on another example | Company-wide adoption |
| Was the result acceptable? | Review against agreed quality criteria | Every future answer will be correct |
| Was time saved? | Task time plus checking and correction time | A universal productivity percentage |
Training does not create permission to use a new tool or share a document. Your organisation’s policies apply throughout the exercises and afterwards.
