Skip to content
Slate and sage impressions meet an ochre field around a terracotta overlap.

Champions and communitiesArticle

How to launch an AI champion network across departments

Map the groups you can support, assign decision owners and rehearse the route to help before launching your AI champion network.

Jump to a section

Launch an AI champion network by agreeing a limited offer of help, the groups it covers and the people responsible for supporting it. Before announcing the network, check that a colleague can find a champion, try an appropriate task and get an unresolved question to someone who can act.

You can begin with a few departments where those conditions are ready. Keep the gaps visible and give each one an owner. An organization-wide invitation needs organization-wide support, or a clear explanation of where the first offer applies.

This guide provides a coverage map and launch checklist. If you are still deciding who should take the role, start with the champion selection guide. For the wider decisions about access, learning and trust, use the people and change guide.

Define the first offer of help

Begin with work that colleagues recognize. For a furniture retailer, the first offer could help warehouse staff summarize unresolved delivery problems for the next shift and help customer-service advisers draft replies about those same deliveries. Both groups need accurate order details, but their outputs and reviewers differ: the warehouse supervisor checks the handover, while the customer-service lead checks what a customer can be promised.

Agree three things with the sponsor:

  • The people and tasks covered. Name the initial groups and what they can ask for help with.
  • The contribution expected from champions. This might be demonstrating a checked example, helping a colleague make an attempt and bringing unresolved questions to a coordinator.
  • The decisions owned elsewhere. Access, information-use rules, specialist quality standards and workload changes need their appropriate owners.

OpenAI Academy's guide to a network of local AI Activators places local champions alongside leaders, sponsors, functional owners, administrators and technical partners. It also allows a limited start without complete organizational coverage. That is vendor guidance, rather than evidence of an ideal network size. The practical implication is to agree who supports the local role before inviting people into it.

Choose a coordinator who can keep those commitments together. In a smaller organization, an existing adoption or learning lead may do this. The coordinator needs access to the sponsor and support owners; asking a keen volunteer to chase every unresolved decision would leave the same gap under a new title.

Map the coverage you can support

A department name does not tell you whether all its colleagues can use the offer. Check working hours, location, language, access to the chosen channel and the kinds of tasks people need help with. Ask the people in those groups where the proposed arrangement would be difficult to use.

Suppose the retailer's warehouse and customer-service teams each have a willing champion with agreed time and a reviewer. A purchasing colleague also volunteers, but the purchasing manager has not agreed who will check supplier emails. A launch map might look like this:

Group and initial taskAvailable helpLaunch condition
Warehouse, summarize delayed or damaged deliveries for the next shift
Named champion, scheduled practice and the warehouse supervisor
Confirm that colleagues on the later shift can attend or use another supported route.
Customer service, draft replies about delayed deliveries
Named champion and a service lead who checks order details and promised dates
Use approved material and confirm the channel is accessible to the invited team.
Purchasing, draft supplier emails about missing goods
Interested volunteer; purchasing reviewer unconfirmed
Keep this group outside the first offer. The coordinator owns resolving the gap before expansion.

The map describes an arrangement to verify, not a staffing formula. One champion might support several small groups with similar work. A single large department may need different arrangements for different tasks or working hours.

For each proposed champion, link the offer to an agreed protected-time allowance. For the colleagues attending practice, confirm how it fits with their delivery commitments. Those are separate capacity decisions.

Connect champions to people who can decide

Champions can explain an example and help someone think through an output. They may also discover a problem they cannot resolve: a missing licence, an unclear data rule, an unsuitable workflow or a manager who cannot release time.

Write down where those questions go and who will respond. A shared channel can make a question visible, but it does not assign responsibility for answering it.

Microsoft's Power Platform champion guidance asks organizers to plan support, sponsorship and communication across organizational boundaries. It describes champions as business representatives rather than a formal support team. The product context is broader than generative AI, but the distinction helps when deciding what colleagues can expect from your network.

A contrasting view comes from Melinda Tan, who advises regulated enterprises on AI adoption. In a public LinkedIn post, she argues that champions cannot fix rollout problems when they lack authority over workflows and resources. Her follow-up promotes her organization's training offer, and the post provides no independently verified sample. The useful challenge is whether your sponsor has assigned the decisions that peer enthusiasm cannot settle.

Give questions a route to help

  1. Try the task

    A colleague tries an approved task and asks for help.

  2. Work with a champion

    The champion helps with the attempt and identifies questions beyond their remit.

  3. Involve the decision owner

    For questions the champion cannot resolve, contact the named owner below.

Close the loop

Agree how the reply reaches the colleague who asked for help.

QuestionNamed decision owner
Access
Administrator
Information use
Guidance owner
Output quality
Task reviewer
Capacity
Manager or sponsor

Ask each owner to acknowledge the route and agree how colleagues will hear what happens next. Where no owner is available, narrow the advertised help or resolve the gap before launch.

Rehearse an exchange across teams

Before the wider invitation, run a small practice session using the proposed channel and an approved example. Include a colleague who did not create the example. In the retailer scenario, ask a customer-service adviser to use the warehouse's handover to explain a delayed sofa delivery. Can they tell whether a date is confirmed or only expected? That distinction may be obvious to warehouse staff and unclear to someone preparing a customer reply.

  1. Find the starting point. Ask the colleague to locate the example, understand its purpose and check whether the task is suitable.
  2. Make an attempt. Let them use appropriate material and explain what they are trying to produce. The champion can help without taking over the entire task.
  3. Check the result. Have the colleague explain what they verified, what remains uncertain and who should review it.
  4. Try a handoff. Use a question beyond the champion's remit. Confirm that the named owner receives it and that the colleague knows where to expect a response.
A colleague points to a document beside her laptop while a peer listens with an open notebook.
Try the example with someone who did not create it, and listen for the questions it leaves unanswered.

Also walk through what happens if access fails or the champion is absent. A rehearsal can end with a smaller offer, a clearer instruction or a different support arrangement. Completing the session is less useful than finding those changes before more people depend on the network.

In its account of a cross-department agency program, Think Big Leaders describes designing the work with the client's existing AI task force and providing office hours for questions arising from live work. The client is unnamed and no independent outcome evaluation is provided. The account illustrates an implementation choice: the people learning had a route back to the program with the problems they encountered.

Announce what is ready

The first invitation should tell colleagues what they can do now. Keep it consistent with the coverage map and the commitments people have accepted.

Before sending it, check that you can state:

  • Which groups and tasks the initial offer covers.
  • Where to find the champion, example and practice opportunity.
  • Which approved tools and material the activity uses.
  • Where urgent or out-of-scope questions belong.
  • What to do when the named champion is unavailable.
  • When the initial arrangement will be reviewed.

Give people a way to ask a question without posting confidential work into a broad channel. An internal example may still need access restrictions. Use the organization's existing information-handling guidance and owners when deciding what can be shared.

Do not advertise an always-available service when the agreement covers a scheduled practice session. People can use a limited offer when its boundaries are clear; an unreliable promise is harder for both learners and champions to manage.

Use the first review to choose what changes

Set the review date before launch and keep its question practical: can people use the offer, and can the network support what it has promised?

Bring the initial coverage map, a small number of attempts and the unresolved questions. Ask:

  • Could colleagues find help?
  • Did the example work in another team's context?
  • Did the right owner handle the blocker?
  • Did the agreed capacity hold?

Then choose the next commitment. You might improve an example, change a session time, add a reviewer, narrow the offer or extend it to a group whose support is now ready. Record the owner of each change. Attendance alone cannot establish that people could apply what they learned.

For your next planning meeting, bring the coverage map and one proposed cross-team rehearsal. Leave with a supported first offer, named decision owners and a date to review the arrangement.

Once the network is running, use the community stewardship guide to follow unanswered questions, keep examples useful and decide what the next gathering should address.

Questions about launching an AI champion network

Does every department need a champion before launch?

No. An AI champion network can start with the departments whose support is ready, provided the invitation clearly names those groups and the help available. A furniture retailer could begin with warehouse handovers and customer-service replies while purchasing waits for someone to review supplier emails. Give that gap an owner and a review date. Use the coverage map to check working hours, access and reviewers before describing the network as available to everyone.

How many champions should the first network have?

Choose the number from the work and people you can support; the sources reviewed here do not establish a universal employee-to-champion ratio. One warehouse champion may cover daytime delivery handovers but be unavailable to the later shift. Check whether that shift has a supported way to practise and ask questions before counting the warehouse as covered. Map those needs, then agree protected time for each champion. Add people where they close an identified gap.

Who should own the network?

Give the network a named coordinator and a sponsor who can resolve priorities and resource decisions. The coordinator maintains coverage, brings support owners together and follows up unanswered requests. Champions help colleagues practise; administrators, task reviewers and managers keep responsibility for their own decisions. For example, a customer-service champion can help an adviser draft a delivery reply, while the service lead decides what date can be promised. Use the decision-owner map to make those routes explicit before launch.

What if the first session attracts only existing enthusiasts?

Treat attendance as information about the first invitation, not proof that the wider team is covered. Ask colleagues who did not attend whether the task was relevant and whether they had access, permission and time to try it. A later-shift warehouse employee may need practice during their own working hours; a customer-service adviser may need a delivery-reply example rather than an internal handover. Change the offer around those findings and rehearse it with someone who did not create the example.

Updated

aiready

A home for your company’s AI community.

Share what works and help each other put AI into practice.

  • Real use cases

  • Practical guides

  • Company policies

  • Shared experience

Explore aiready
Explore the blog