An AI champion is an employee who, alongside their normal job, answers colleagues' AI questions, collects what works, and reports what breaks to whoever owns the AI programme. In a team using Copilot, ChatGPT Enterprise or Claude, they are the first person people ask before they ask IT.
In practice the role is seven recurring tasks that add up to about 15 hours a month, or roughly half a day a week, once the training is over. Choose the person colleagues already trust rather than the keenest volunteer, put those hours in writing, and the role lasts. Leave it as goodwill and it fades within a quarter.
One clarification, because search results blur the two. This post is not about the UK government's AI Champions, the industry figures HM Treasury appointed in January 2026 to speed up AI adoption across financial services. It is about the person inside your own organisation that somebody has just asked you to appoint, with no definition beyond the phrase.
Most champion programmes start with an all-hands email asking for volunteers. Two of the three decisions that sink them have already been made by the end of that email.
What counts as part of the AI champion role?
A task belongs in the AI champion role only if it is something the champion does, not something they are, and only if it has a time cost you can write down. "Passionate about AI" fails both tests. "Runs a 45-minute clinic every other Thursday" passes.
The test matters because most role descriptions are lists of qualities. Microsoft's Copilot champion charter template lists expected time commitment as something to define, then leaves the number blank. A role with no hours attached cannot be protected from the rest of the person's job, and the rest of the job always wins.
Every item below therefore carries three things: what the champion does, a rough monthly budget in hours, and what happens if nobody does it. The hours are a planning estimate for a team of 20 to 60 people after training has finished, not a measured benchmark. Expect nearer a day a week during the first six weeks of a rollout across a 100-person team.
The 7 things an AI champion actually does
1. Answers the questions that would otherwise go nowhere
The champion is the named first stop for small questions: why Copilot cannot see a SharePoint folder, whether a client contract can go into Claude, which model to pick for a summary. Most of these take five minutes to answer and would never be raised with IT or a manager.
Time: around 4 hours a month, front-loaded into the weeks after training.
If nobody does it: the question goes unasked, the person tries once, gets a poor result, and concludes the tool does not work for their job. That conclusion is rarely revisited.
2. Collects the use cases that worked, so others can reuse them
When someone in finance gets a reliable month-end variance commentary out of ChatGPT Enterprise, the champion writes down the prompt, the inputs and the check they ran on the output. One shared page, organised by task rather than by tool.
Time: around 2 hours a month.
If nobody does it: every good use case lives in one person's chat history. The team keeps reinventing the same prompt, and when that person leaves, the method leaves with them.
3. Acts as the fast route for a tool request, not a blocker
When someone wants to try a new tool or connector, the champion checks it against the approved tools list, answers in days rather than weeks, and passes anything new to whoever owns the decision with the context already gathered.
Time: around 1 hour a month in most teams, more in technical ones.
If nobody does it: requests stall in a queue, and people move the work onto personal accounts where nobody can see it. A slow approval route is one of the most reliable ways to create shadow AI use.
4. Translates between the people who build and the people who never will
Every team now holds a few people building automations and agents, and a larger group who will never build anything and should not need to. The champion explains to the first group what the second group actually needs, and to the second group what is now possible without a ticket.
This gap is wider than most budget holders think. In ivee's research with 500 UK AI decision makers, surveyed in August 2026, 75% said either that they have no idea, or that fewer than 1% of their workforce knows how to build and deploy an AI automation.
Time: around 2 hours a month, mostly in conversations that would happen anyway.
If nobody does it: the builders build things nobody uses, and everyone else keeps doing the work by hand.
5. Feeds real failures back to whoever owns the programme
The champion keeps a short running note of what went wrong: the hallucinated figure that nearly reached a client, the prompt that worked in testing and failed on real data, the policy rule nobody understands. Once a month, it goes to the programme owner in a 30-minute call.
Time: around 1 hour a month.
If nobody does it: the programme owner hears only success stories, because those are the ones people volunteer. The next round of training repeats the gaps of the last one.
6. Runs what happens after the training day
Training gives a team the method. The fortnight after decides whether anyone uses it. The champion runs a short clinic every other week where people bring a real task, works through stuck cases one to one, and keeps the examples from the training current.
Time: around 4 hours a month, including preparation.
If nobody does it: usage peaks in the week after training and drifts back towards zero. This is the most common way an otherwise good programme fails, and it is the task most often left off the role description.
7. Keeps the rules usable
The champion knows the AI policy well enough to answer "can I put this in?" without sending people to a 12-page PDF. In practice that means one shared red, amber and green list of data types, explained in the team's own terms, and updated when the policy changes.
Time: around 1 hour a month.
If nobody does it: people either avoid the tools entirely or ignore the policy. Both are reasonable responses to rules nobody can apply in the moment.
How do you choose an AI champion?
Choose the person colleagues already go to for help, not the person most excited about AI. Credibility transfers between colleagues and enthusiasm does not. A sceptical account manager will try a method because the colleague who always gives straight answers showed it to them, not because the office early adopter did.
The default route works against this. GitHub's playbook for internal AI advocates recommends finding champions by asking for volunteers, on the grounds that the most motivated people will raise their hands. That is a fair way to find motivation, and motivation does matter. It also selects for the people least representative of the colleagues they are meant to win over.
Two further rules hold up well:
Spread champions across functions. Concentrating them in engineering or marketing, where adoption is already easy, leaves finance, operations and HR with nobody. One per team of roughly 20 to 40 people is a sensible starting ratio, adjusted to how the work is actually split.
Check capability, not self-description. Identifying the likely champion is one of the baseline items in what to assess before you train a team in AI, because a tested result is more reliable than a volunteer's own view of their skills.
How much time does an AI champion need?
An AI champion needs about 15 hours a month in steady state, which is roughly half a day a week, and nearer a full day a week during the first six weeks of a rollout. Those hours need to come out of the person's existing workload rather than sit on top of it.
The seven tasks above sum to that figure. If the role description asks for all seven and allocates two hours a month, it is describing a job nobody can do, and the first thing to go will be the clinic in item 6, which is the one that matters most.
Write the hours into the person's objectives and agree them with their line manager before announcing the role. The line manager is the person most likely to reclaim the time when a deadline lands, so their agreement is the commitment that counts.
What should you give your AI champions?
Give AI champions three things: a named number of hours, early access to tools and changes, and a direct line to someone who can make a decision. Badges, a Slack channel and a certificate are pleasant extras, not substitutes.
Named hours, agreed with their manager and written into objectives, so the role survives a busy quarter.
Early access to new tools, licence changes and policy updates, so they are never the last to hear something their colleagues are asking about.
A direct line to a decision-maker, usually the programme owner or the COO, so a tool request or a policy question can be settled in days.
Without the third, the champion becomes a messenger who collects problems and cannot resolve any of them. That wears people out faster than the workload does.
When does an AI champion network not work?
An AI champion network does not work as a substitute for a mandate, for allocated time, or for changing what people are measured on. Champions make a programme easier to adopt. They cannot make it matter if leadership has not said it matters.
Four limits are worth stating plainly:
Without allocated time, it does not work at all. A champion role run on goodwill fades within a quarter, and the programme loses its only internal owner without anyone noticing.
Under about 30 people, a network is overhead. One named person with a few hours a month and a direct line to the founder does the job. A network, a charter and a monthly forum are structure for an organisation that does not yet need it.
Measures beat advocacy. If people are still judged on output that AI does not touch, no champion will persuade them to spend time learning it.
The champion will leave. They are, by definition, one of your more employable people. The shared use-case page from item 2 and a named deputy are the plan for that day. The same key-person risk is why the decision on running AI training in-house or with an external provider usually turns on this one person.
Treat the champion as the person who keeps a programme running between decisions, not the person who makes it succeed. Define the seven tasks, cost them, protect the hours and plan the handover, and the role becomes something a good colleague can say yes to.
To see how champions fit around the training itself, read how ivee's AI services work, from strategy days to team programmes.
