Teach the team to use AI on the work they actually do.

Different skills for preparing, reviewing and managing work.

A person drafting an answer needs to supply useful context. A reviewer needs to spot missing facts, weak sources and unsupported conclusions. A manager needs to know who receives exceptions and when a workflow should be paused. In a small company one person may hold several roles, but the skills remain different.

Start with the tools the organisation has approved and the tasks people are permitted to perform. Use synthetic examples or data authorised for training. Participants should not need to upload confidential documents into personal accounts simply to complete an exercise. For multilingual groups, agree the working language and examples before the programme.

Example workshop: review a support response.

In this illustrative exercise, participants receive a customer request and a short procedure. They prepare a response, then compare it with an answer containing a plausible detail that is deliberately absent from the source. Their task is to identify the defect and improve the review method.

A possible structure for the programme.

1. 01

   Establish the starting point

   Collect frequent tasks, concerns and tools already in use. An initial exercise helps distinguish unfamiliarity with the interface from difficulty judging the output.

2. 02

   Practise a few relevant tasks

   Use role-specific exercises and reusable task instructions. Participants compare different results for the same input and discuss what makes an output acceptable.

3. 03

   Rehearse failures

   Introduce incomplete sources, conflicting instructions and out-of-scope requests. Practise explaining a limitation, inspecting the source and handing the case to its owner.

4. 04

   Support early use

   The agreed scope can include a later review of examples collected by the team, revisions to task sheets and an internal owner for recurring questions.

Assess independent judgement, not just attendance.

Activated accounts and positive feedback at the end of a session do not establish improved work. A useful assessment gives participants an unfamiliar example and checks whether they select a relevant source, identify an error and complete the handoff without continuous assistance.

Training should leave concise task sheets covering permitted work, approved data, review criteria, failure examples and escalation owners. Update those materials when a tool or procedure changes. One session should not be assumed to solve access, adoption and process ownership at once.

Choose training that fits the operating environment.

NIST places clear responsibilities and personnel preparation within AI risk management. It provides organisational context, not a certification of the training. Any formal, regulatory or professional training requirement needs its own agreed scope and the relevant qualified owner.

• Exercises correspond to tasks participants genuinely perform.
• Data, tool and process owners approve the scope before the workshop.
• Practical evaluation and error discussion accompany feature explanations.
• Materials can be revised and have an owner after delivery.

Practical questions