Many AI rollouts get champions shortly after the seats are purchased. Their names go into a slide. They get an invite to a Teams channel, maybe a short training session, and an informal request to help their department get excited.
Six months later, leadership asks how adoption is going. The champions have hosted a few office hours. Usage is uneven. The people who were already curious are further ahead. Everyone else has learned that opening the tool is optional.
The program had people. It never had a job.
An AI champion program works when each champion has one recurring workflow to improve, manager-approved time to improve it, and someone who can say whether the result is useful. Enthusiasm is not enough. The champion needs an assignment.
Seat activation is a poor job description
Most champion programs start with the visible work: training sessions, tips in chat, a monthly show-and-tell, answers to simple questions. Those activities help people discover a tool. They do not explain why a team should change a recurring piece of work around it.
Microsoft’s own guidance separates champions from the people who own adoption, service reliability, and subject-matter quality. It describes champions as business representatives who support peers and bring feedback back to the program, rather than a substitute for the support desk or the owner of a production service.1 That boundary is useful. A champion can expose where the work is ready to change. Someone still needs authority over the workflow and its result.
The common failure comes from treating an active user as an adoption plan. The person who has figured out Claude for their own work can show the team a better way of drafting a brief or preparing a report. They cannot quietly take responsibility for every team’s tool choice, training, data rules, and support queue. That assignment turns the person producing unusual output into an unpaid internal help desk.
The Driving Adoption guide covers the larger diagnosis. Tools go idle because the tool, workflow, manager signal, or people are wrong. A champion program belongs in the workflow part of that diagnosis. It gives the person who found a better method time and backing to make it visible and reusable.
The champion owns a handoff, not a department
The right first assignment is small and concrete. A marketing champion might take the weekly competitor brief: collect the same sources, have AI produce the first pass, add the checks that catch a bad claim, and leave the final call with the marketing lead. A finance champion might use AI for a first pass through invoice exceptions while leaving approvals with the person who knows the contracts.
Both assignments have an output, a review point, and a person receiving the work. That gives the champion a finish line. “Help the team use AI better” has none.
The job description can fit on a page.
| Part | What to name | Why it matters |
|---|---|---|
| Mandate | One recurring output that should change in the next quarter | The role begins with work, not enthusiasm. |
| Business owner | The person who accepts the finished result and owns exceptions | The champion can improve the method without inheriting another leader’s decision. |
| Protected time | The manager-approved time for testing, documenting, and teaching the workflow | Work done only after hours disappears when the quarter gets busy. |
| Guardrails | The approved tools, data boundary, and point where a person must review | A useful shortcut should not create a quiet security or quality problem. |
| Measure | A before-and-after signal such as cycle time, rework, queue age, or completed work | Active seats measure availability. The changed output measures value. |
The role should also state the decision the champion may make. They can recommend a workflow, document the method, and ask for an experiment. The business owner decides whether the output is good enough to use. Security and IT own the controls around access and data. A champion who owns every one of those choices will either stall or build an ungoverned side system.
It keeps the champion in the workflow and leaves decisions about the output, access, and data with the accountable people. A shared Claude Skill follows the same rule: the instructions can travel, but an owner and reviewer stay attached to the result.
Work artifacts reveal the strongest champions
Work artifacts and usage patterns reveal candidates sooner than a sign-up form. The person who made the Monday report arrive early, turned a messy intake into a clean queue, or keeps finding a better first draft has evidence worth inspecting. Recognizing leverage explains how to find that evidence without confusing message volume for output.
Interest still matters. A person can be highly effective alone and have no appetite for documenting or teaching the method. The useful conversation asks which recurring task they changed, where the result still needs judgment, and what would let a colleague use the method without a private lesson. Those questions also make a better screen when you are hiring for AI skills.
Manager support is the other test. Microsoft recommends setting the expected commitment, communicating the requirements, and getting manager approval where it is needed.2 A director should read that as a budget decision. If the champion has no protected time, the company has decided the role belongs after the day job. The program will produce a few helpful messages and no durable change.
The first result needs to be narrow enough to inspect
The first workflow should recur often enough to matter and have a reviewer who can tell a good result from a bad one. The verification test is a useful filter here. A workflow with a fast check, clear source material, and a contained error path can move quickly. A high-stakes decision with no shared standard for quality needs more preparation before a champion tries to package it.
The champion’s first review with leadership should be short. What did this output look like before? What changed in the method? Who checks it now? What did the team gain: time, fewer corrections, a shorter queue, or a better decision? If the answers are clear, another team has something worth adopting. If they are fuzzy, the program has learned where more workflow design is needed.
That is a much better conversation than how many people attended a lunch-and-learn. Training attendance reports an event. A changed recurring output gives leadership a reason to continue funding the role.
The job description is real when a leader can name the workflow, the business owner, the review point, and the measure. Then the champion is changing a recurring piece of work instead of carrying a new tool around the building.
Footnotes
-
Microsoft, “Define roles, responsibilities, and decision rights”, updated July 2026. It separates champions, adoption leads, service owners, and domain experts. Last verified August 12, 2026. ↩
-
Microsoft, “Support Power Platform champions in your organization”, updated May 2025. Its program guidance covers expected commitment, clear requirements, and manager approval. Last verified August 12, 2026. ↩