An AI Training Plan for One Business Team

A company can give everyone an AI course and still leave managers unsure whether staff can use the tool safely in their jobs. Start with one business team and one task that already has a clear owner and quality standard. The aim of training is a reliable way to do that task, including when to decline an AI answer and ask a person.

NIST’s AI Risk Management Framework calls for AI risk training consistent with people’s roles. That supports role-specific practice; it does not prescribe a four-week curriculum. The schedule below is an editorial template to adapt to your workflow and risk.

Week 1: define the work and the boundary

Pick a task such as drafting an internal brief from approved material. The manager writes down what goes in, what an acceptable output contains, who checks it, and where the final version belongs. A security or data owner names information that cannot enter the tool. Give staff two real, de-identified cases and one intentionally unsuitable case. They should be able to explain why the unsuitable one must stay out.

Week 2: practice with a visible review step

Show one complete example: source material, instruction to the tool, draft, human checks, corrections, and accepted result. Then ask each participant to work through cases with the same rubric. The rubric should cover factual support, missing context, restricted information, and the actual business requirement. “Looks good” is not a review standard. A reviewer records where a draft failed and how long the correction took.

Week 3: use the workflow under supervision

Let the team apply the method to a limited set of eligible work while a named reviewer still approves each result. Track eligible cases, attempted cases, accepted cases, repairs, and material errors. Keep the current process available for exceptions. Invite staff to report prompts that fail, source gaps, and reasons they choose not to use the tool; those are clues to improve the workflow, not marks against the employee.

Week 4: decide whether the team is ready

Ask people to handle one normal case and one failure case without coaching. Can they spot an unsupported claim, trace it to a source, correct it, and escalate when they cannot? Have the manager compare accepted work and review effort with the old process. If the team cannot explain the boundary or quality falls, keep supervision and revise training. If it passes, expand one task or cohort at a time with a continuing owner.

NIST’s AI RMF Playbook suggests training that addresses system limits and possible negative impacts. Put those limits into the exercises, not merely a slide. The manager owns the operational decision; attendance records only show who entered the room. Our workflow evaluation lesson can help turn the practice cases into a reusable check.

The next cohort

Before teaching another team, write down the cases that failed, the rubric changes, and the reviewer time. Reuse the materials only where the task and data boundary truly match. Training scales when people can repeat the decision and recover from a bad output, not when the completion percentage reaches 100.

Useful next steps

• Start with AI for absolute beginners — Give new users a shared starting point.
• Learn workflow evaluation — Turn practice cases into a repeatable test.
• Use the AI pilot scorecard — Make the scale decision from accepted work.