AI training needs analysis: stop guessing what your team needs

The quick answer

An AI training needs analysis compares what people need to do at work with what they can demonstrate today. Start with tasks, observe a small sample of work, and separate a learning gap from a missing permission, unclear process or unsuitable tool. The output should be a prioritised training brief, not a league table of confidence scores.

Use this approach when someone asks for “AI training for everyone” and you need to turn that request into something you can commission. The worksheet below is a proposed working method, not a validated assessment instrument.

Ask about the last real task

Interview people from the roles you expect to train. Include someone who is enthusiastic, someone who is sceptical and someone responsible for reviewing the work. Ask them to describe a recent task rather than speculate about everything AI might do.

Use these six questions:

1. What were you trying to produce, and who used it?
2. Which steps took the most effort?
3. Where did you need judgement or information from another person?
4. What would an unacceptable result look like?
5. What tools and information are you allowed to use?
6. Could you show how you would check an AI-assisted version?

Do not collect confidential examples just to make the interview feel realistic. A redacted or synthetic version can reveal the same learning need. Ask the relevant owner before moving any work material into a training environment.

Copy this evidence sheet

Create one entry for each task, with these fields:

• Role and task: who does the work and what they produce.
• Current evidence: an observed attempt, reviewed output or structured conversation.
• Desired behaviour: the specific action a learner should demonstrate.
• Gap: what is missing today.
• Likely cause: knowledge, practice, process, access or tool fit.
• Practice activity: the task learners will attempt in training.
• Reviewer: who can recognise a good result.
• Follow-up: when you will check a fresh example at work.

For a hypothetical account team, the desired behaviour might be: “Draft a client follow-up using approved notes, identify unsupported claims and obtain the normal review before sending.” That is teachable and observable. “Become AI confident” is much harder to assess.

Decide what deserves training first

For each gap, discuss how often the task occurs, the consequence of getting it wrong and whether the team has an approved environment in which to practise. Start where the need is meaningful and the conditions for practice exist.

A blocked account does not need a prompting workshop. A policy question needs an accountable decision. Someone who cannot identify invented facts needs guided checking practice. Keep those actions separate in your plan, even when they belong to the same initiative.

Record who you spoke to and whose perspective is missing. A handful of interviews can uncover useful patterns, but they do not establish the needs of an entire workforce. Validate the brief with the managers and reviewers who will support the learners.

Turn the findings into a brief

Write one paragraph describing the audience, the task, the current gap and the evidence you want after training. Add the constraints: available time, approved tools, accessibility requirements and review responsibilities. Give the provider an example of acceptable work where you can safely do so.

Before commissioning, read how to commission workforce AI training. If you need to map different expectations across roles, use the AI skills matrix template.

Start smaller than a workforce survey

Try the question set with one team and review whether it produces decisions you can act on. Remove questions that collect interesting information without changing the plan. Repeat with other roles when the method is useful.

You can begin with the Experrt learning check, which asks for your name and email before showing your self-reported priorities. It is a conversation starter, not a workforce diagnosis. Explore sponsoring an AI literacy programme, or talk to Experrt about turning the findings into a programme.