AI champions need a job description, not another badge

Define what champions are there to do

An AI champion can help colleagues discover useful workflows, practise with approved tools and bring recurring problems to the right owner. That role needs boundaries. Being enthusiastic about AI does not make someone the organisation’s security, legal or technical authority.

Write the remit before recruiting volunteers. Explain what support champions provide, what decisions remain with managers and where questions should be escalated. Without that agreement, a champion network can become an informal help desk with no capacity or authority to resolve the problems it receives.

Copy this champion charter

Use the following as a proposed starting point and adapt it with your managers:

• Purpose: help colleagues practise suitable AI-assisted tasks in the approved environment.
• Activities: host a short practice session, maintain agreed examples and collect feedback.
• Boundaries: do not approve new tools, data uses or exceptions to policy.
• Time: agree a specific allocation with the manager and review whether it is sufficient.
• Support: name a training lead and the owners of technical and policy questions.
• Evidence: record useful workflows, unresolved barriers and examples that failed.
• Review: revisit the role after the initial pilot and adjust or stop activities that are not useful.

Make the charter visible to colleagues. It should help someone understand what they can ask a champion and when they need another route.

Recruit for more than enthusiasm

Include people who know the day-to-day work and can explain it patiently. Look for colleagues who are comfortable saying “I don’t know” and who will check an answer before sharing it. A confident demonstration is less valuable if it encourages others to trust unverified outputs.

Aim for coverage of relevant roles and working patterns. A network drawn only from head office may miss the access, timing or process constraints faced by other teams. Ask managers where peer support would actually help rather than choosing a fixed number of champions for every department.

Participation should have a clear agreement about time and responsibilities. Avoid making unpaid extra effort the hidden foundation of the programme.

Teach a repeatable support pattern

Give champions a small set of approved exercises and a way to run them. A useful session starts with the task, checks the information boundary, lets people attempt the work and finishes with a review of the output.

Teach the escalation pattern alongside the exercise. If someone asks whether a new dataset can be uploaded, the champion should know who decides. If an output is unreliable, the champion should help the colleague check it and record the issue, not improvise a reassuring explanation.

Use the AI prompt template as a practice aid, with the review steps kept attached.

Make managers part of the programme

Managers choose priorities, make time for practice and decide whether a workflow belongs in normal operations. Champions can support those decisions but should not be expected to replace them. Agree which team task the programme is trying to improve before measuring the number of sessions held.

Read managers, not just champions, drive AI adoption for the management side of the programme.

Review the network by usefulness

After the pilot, ask what colleagues can now do, which barriers were resolved and which questions still lack an owner. Keep attendance as an activity measure, alongside evidence of changed practice. Do not treat the number of enthusiastic posts as proof of business impact.

Explore leading an AI-ready team or contact Experrt about preparing champions and managers together. Start with a clear role, one useful workflow and enough support to make both sustainable.