An AI champion should be able to teach one approved workflow, check its output and recognize when a question needs another owner before teaching colleagues independently. Prepare them for that specific assignment: the work, the tool, the company's rules and the learner's starting point. Then rehearse the teaching with someone who can give useful feedback.
This guide is for an adoption lead preparing an already selected peer coach. If you are still choosing volunteers, use the power user and AI champion guide first. Here, the decision is what the person needs to learn before their first supported teaching assignment.
Start with the work they will teach
A champion's preparation needs a task that colleagues recognize. “Teach AI” leaves too many decisions unresolved: which information can be supplied, what a usable result looks like, and who checks it before it affects a customer or colleague.
Choose one workflow with its work owner. Build a small learning pack around it:
- Task and purpose. Name the draft or other output the learner will produce, and who uses it.
- Permitted inputs. Supply an approved practice example and the tool colleagues are allowed to use.
- Quality checks. Identify the facts, requirements and omissions the learner must compare with the source material.
- Boundaries. State what the tool cannot decide and who takes unresolved questions.
- Teaching support. Give the champion a checked example, a colleague to rehearse with and a person to ask when the instructions are unclear.
OpenAI Academy's champion deployment guide combines foundations such as clear instructions, context, checking responses and responsible use with local support and applied examples. Its guidance distinguishes course participation from what people apply afterward. It is vendor guidance, rather than evidence that a particular course makes someone ready to teach your workflow.
Use foundation learning to fill gaps, then work through the local example. A champion should be able to explain why a particular input is needed and why an output would be rejected. Reproducing the trainer's prompt is a smaller achievement.
Practise finding the mistake
Suppose Jamie Lee is the reception team's AI champion at a hotel. Jamie will help Alex Morgan draft replies to guest requests to change booking dates. They use an approved tool and fictional training records prepared by the reservations manager.
The practice booking runs from July 10 to 12. The guest asks for July 17 to 19, but the supplied availability card says those dates have not been confirmed. A useful draft acknowledges the request and leaves confirmation with reservations. A polished reply promising a room would be wrong even if it copied the dates correctly.
Jamie needs to learn the check before teaching the prompt. During preparation, ask Jamie to locate the evidence for each important statement in a sample draft. Include a deliberately wrong draft that promises availability, so the exercise does not depend on the model making a convenient mistake on demand.
A public August 2026 Reddit discussion illustrates the problem with unexplained output. A Reddit user in a self-described data-security-focused role reported a coworker sending Copilot paragraphs that did not answer the question and generating scripts they could not explain. The account is unverified and does not establish how common this is. It does make a useful preparation question concrete: can the champion explain the work, rather than pass along an AI answer?
Have Jamie practise correcting the promise, explaining why it was unsupported and identifying what still needs confirmation. The staff proficiency guide covers broader assessment of work people can complete and explain. Champion preparation adds a teaching responsibility: helping someone else perform the check.
Learn to leave the learner in control
Knowing the task and teaching it are separate skills. Rehearse with a colleague who is less familiar with the tool and willing to say when an explanation loses them. The champion needs practice asking a useful question, waiting for an answer and adjusting the explanation without taking over.
AI transformation consultant Sara Maldon described a mismatch between enthusiastic volunteers and learners after conversations with other transformation leads. Volunteers may want early access to tools, while learners need a slower pace and room for basic questions. She described coaches becoming bored as learners get lost. This is a consultant's account, not an independently evaluated programme, but it exposes a teaching habit worth practising.
In the hotel rehearsal, Alex might say the draft looks fine. Jamie can ask, “Where does the practice record confirm a room for July 17 to 19?” Alex then compares the statement with the availability card. If Jamie immediately rewrites the reply, the observer learns much less about whether Alex understood the problem.
Give feedback on the teaching as well as the result:
- Explanation. Did Jamie explain an unfamiliar term?
- Pace. Did Alex get time to try?
- Understanding. Could Alex describe the check afterward?
A learner's confusion can reveal a weak explanation or an unclear practice pack; it is not automatically evidence that the learner needs more motivation.
Learn the referral route before promising help
A champion should know whom to contact for questions beyond the assignment, and how to describe the issue so that person can act. Microsoft's Power Platform champion guidance places champions in the business community and distinguishes them from a formal support team. That broader platform guidance is a useful boundary for a peer coach too.
NIST's AI Risk Management Framework Playbook, Govern 2.2 connects training to people's duties and includes documented escalation paths. The playbook offers voluntary suggestions, not a required training checklist. The implication for your preparation is practical: teaching more tool skills will not resolve a missing decision owner.
For the hotel example, assume the practice policy reserves fee exceptions for the reservations manager. When Alex asks whether the fee can be waived, Jamie should identify that boundary and help formulate the question. Jamie can summarize the requested date change and the unresolved exception without claiming permission to waive it.
Prepare a short support note using only information appropriate for that channel:
- Question. What decision or help does the learner need?
- Checks completed. What has the learner already compared or tried?
- Unknowns. What still needs confirmation?
- Owner. Who should receive the question?
If nobody owns the decision, pause teaching that part of the workflow while the adoption lead and work owner resolve it.
Rehearse the assignment before the first session
Once the champion has practised the task, checking, coaching and referral, bring those capabilities together. Use a work owner or experienced facilitator as the observer and a colleague as the learner. Tell both people what you are looking for; avoid turning a preparation session into a surprise exam.
A preparation rehearsal for one workflow
Teach the ordinary case
The learner drafts a reply from the hotel practice pack. Observe how the champion explains the purpose and helps the learner check the result.
Discuss a wrong draft
Give the learner a prepared reply that promises unconfirmed availability. Observe whether the champion helps them find and explain the unsupported statement.
Handle a boundary question
The learner asks for a fee exception. Observe whether the champion identifies the decision owner and prepares a useful referral.
Agree the next supported assignment
Record what the champion handled independently, what help they needed and the preparation to complete before teaching colleagues.
Keep a short observation record with the task, tool, practice materials, help supplied and any unresolved issue. Record the learner's explanation too: a champion doing every correction themselves can leave a good-looking draft without demonstrating useful coaching.
Match the next assignment to the gap
Use the rehearsal to choose the next preparation step. A single label such as “needs more AI training” hides whether the difficulty was task knowledge, verification, teaching or company support.
| What the rehearsal showed | Preparation to arrange | Teaching scope for now |
|---|---|---|
Missed an unsupported statement | Practise comparing drafts with source records and explaining corrections | Rehearse again before teaching independently |
Found the error but took over the learner's work | Practise questioning, pacing and letting the learner retry | Co-teach with an experienced facilitator |
Could not identify who decides an exception | Confirm the policy boundary and a reachable owner | Defer the unresolved part of the session |
Taught the task and its checks, and referred the exception appropriately | Give feedback and arrange a supported first session | Teach this workflow with the agreed support route |
Keep the conclusion specific. Being ready to teach date-change reply drafting does not establish readiness to teach hotel pricing, automate reservations or advise on every AI tool. Revisit preparation when the workflow, tool or relevant policy changes, or when learner questions expose a new gap.
Agree time for preparation and feedback with the champion's manager using the protected-time guide. Your next step is to select one workflow, prepare its practice pack with the work owner and book the rehearsal. The people and change guide connects this assignment with the wider learning and support programme.
Questions about preparing AI champions
Is a course certificate enough to start teaching?
A certificate can show that someone completed a course, but you still need evidence that they can teach your workflow and its checks. Ask the champion to use the approved practice materials, explain an important correction and help a learner make the check. Include a question outside their authority, such as the hotel example's fee exception. Use the preparation rehearsal to decide the first supported assignment, rather than treating course completion as permission to teach every task.
Does a champion need to know how to code?
They need the technical knowledge required by the workflow they will teach. Coaching reception staff to draft and check a booking reply does not require writing software. Teaching colleagues to generate or change a script does require someone who can understand, review and explain that work, with the relevant technical owner involved. Define the output and its checks in the learning pack. Do not extend a successful drafting session into an unsupported coding or automation assignment.
How much preparation time should we allow?
Estimate time from the champion's gaps and the assignment, rather than choosing a universal number of course hours. Include learning the approved tool, preparing the practice material, rehearsing with a colleague and using the feedback. Someone who knows hotel reservations may need coaching practice; someone new to the workflow also needs help from its owner. Agree which existing work moves to make that time available. The protected-time guide provides a planning method, then you can review the estimate after the rehearsal.
What if the champion struggles in the rehearsal?
Identify the specific gap and arrange practice or support before expanding the assignment. A missed availability check calls for verification practice; taking over Alex's draft calls for coaching practice. A missing fee-exception owner requires the organization to resolve the boundary. Record what help was needed, provide it and rehearse the affected part again. The next-assignment table helps distinguish further practice, co-teaching and deferral without turning one attempt into a judgment about the person's overall potential.
