How to Build AI Champions in Your Organisation

By James Cotton, Founder, iO-Sphere

AI adoption happens one person at a time. Someone decides to draft this week's report with an assistant, or to let the model do a reconciliation they used to do by hand. An AI champion is the person who does that first, and whom colleagues copy. A champions programme is the deliberate way a company grows those people and helps their way of working catch on.

Most programmes never manage it. They name an enthusiast, give them a title, and wait. Nothing changes.

What an AI champion is, and why you need to plan for it

An AI champion can do real work with AI, and has enough standing that colleagues follow their lead. They can sit with a peer and say, "here is how I did it, on our data, in our tools," and be believed, because they have done it. The trainer who ran the workshop does not know your data or your Tuesday. A champion does.

Plan for it, because capability is where AI stalls, not tooling. MIT NANDA found that 95% of generative-AI pilots returned no measurable profit or loss, and put the blame on learning rather than the technology ("The GenAI Divide: State of AI in Business 2025", fieldwork January to June 2025). The tools are already on the desk. What most companies lack is people who can use them well, and that is what a champions programme builds.

One limit to be clear about: a champion multiplies the capability you build, but does not replace building it. Lean on one enthusiast instead of training the team and you get an overloaded person and a team that never moves.

How to pick the right people as champions

Pick champions by pull, not rank. The person colleagues already go to for help is where a new way of working spreads from, whatever their job title. Rank tells you who signs things off. Pull tells you who others actually follow.

Look for the person in each team who has both credibility and curiosity. Three quick tests:

• Do people already ask them how to do things? Trust is the thing you cannot train, so start where it already exists.
• Are they curious about the work, not just the tool? Someone who loves the demo but never changes how they work makes a weak champion.
• Are they close to the team's regular tasks? That gives them something real to practise on and something real to show.

Fit comes down to the work someone does. Data and AI run through most jobs now, so more of your people can be champions than the org chart suggests.

The mix that works: a few deep, many fluent

A champions programme has two layers. First, a foundation of working fluency for the whole team, so everyone can use the company's AI tools and knows when to pull in help. Then, on top, one or two people go deep, trained on real work over months until they can handle the hard cases and coach the rest. You give the foundation to everyone; the depth is what a couple of people carry for the team.

Here is that on a team of thirty. The whole team takes a short fluency course, so everyone can use the tools and knows when to pull in help. One or two go deep over a year or so, learning on the team's own reports and reconciliations. When the finance lead needs a report reshaped, it goes to a champion, who has done it before on this team's data. They solve it, show how they did it, and because everyone else has the foundation, a couple of colleagues pick up the approach the next week.

Get the balance wrong and you can see it. Train everyone the same shallow amount and no one is deep enough to help a colleague with a real problem, so the early enthusiasm fades and nothing changes. Train a few deeply but leave everyone else unable to follow, and every question lands on the champion, who burns out in a few months while the rest of the team stays where it was. You need both: the depth to solve a hard problem, and the fluency to spread the answer.

What to teach each group

People get good at AI by doing the work, coached by someone who has done the job, on the tasks they face. An hour-long "intro to AI" video tells people AI exists. It does not help anyone do their job differently on Monday.

When a team's own systems are too sensitive to practise on, we build the practice from realistic case studies and simulations (our Prism method), so people rehearse true-to-life work without going near anything live.

The deep few learn the whole loop: frame the problem, get and question the data, build something with AI, and judge whether the answer holds up, all on the team's own work. Everyone else learns to use the assistants well, to tell when AI is reliable and when it is not, and to move work to and from a champion. Neither group needs a lecture on how the models work. Both need practice.

Give champions a mandate and protected time

Champions succeed when the company backs them, and the backing that matters most is time. Put the hours in the calendar, have leadership defend them openly, and protect them the way an apprentice's off-the-job hours are protected. With real hours, a champion can coach, answer questions, and work out the next thing worth sharing.

A title with no protected time does the reverse. It hands your most motivated person all of the expectation and none of the time, and their motivation turns to frustration fast. If you protect one thing for a champion, protect the hours.

Champions spread a way of working by doing it where colleagues can watch, then letting them copy it. Set a regular rhythm: coached sessions where champions work on real problems, then a session where they show their team. A colleague saying "here is how I did it on our data" carries further than any outside trainer, because it closes the gap between the course and the actual job.

Measure adoption, not attendance

Course completions and headcount tell you people showed up. They tell you nothing about whether the work changed. Judge the programme by who is using AI on Monday morning, and what they now do differently.

The signals that matter, roughly in order:

• Tasks that changed. Specific jobs now done with AI that were not before. This is the whole point; everything else stands in for it.
• Work moving to and from champions. Colleagues bringing real problems to a champion, and the champion handing back a solved approach. That traffic is the network forming.
• Time saved or quality gained on regular work. Measured on the team's actual tasks, not a general productivity claim.
• New habits starting away from the champion. Someone who is not a champion picking up a technique because a peer showed them. That is the practice spreading on its own, which is the goal.

Take a quick read of where each team stands today before you start, so you can see the change rather than claim it. Licences bought and videos watched will flatter a board and mean nothing.

When a champions programme is the wrong fit

This approach fits most employers trying to make AI stick, team by team. Here is when to look elsewhere first.

• If the real blocker is your data or systems, fix that first. RAND put the AI-project failure rate above 80%, about twice the rate of other IT projects, and blamed a poorly understood problem most often (report RRA2680-1, August 2024); bad data usually sits under that. If your teams cannot get clean, trusted data to their desks, champions will not fix it. Sort the pipes, then build the people.
• If you need light fluency across hundreds of people in weeks, this is the wrong shape. Coached depth moves at the speed of practice. For fast, basic confidence across a large team, our short fluency courses do that directly; run those first, then come back to depth once the basics are in.
• If you are a small employer with no levy pot and a tight budget, add up the overhead honestly. The apprenticeship route costs more than fees: protected hours, an end-point assessment, months of managed time. At small scale that can outweigh the funding, so one of our self-funded short courses may fit better than the full programme. Real capability is still worth paying for; just size it to your company.

Other routes: apprenticeship units and vendor training

Two nearby options come up a lot. We do not deliver either, so here is where each fits.

Short apprenticeship units arrived in April 2026. The Growth & Skills Levy (formerly the Apprenticeship Levy) now funds standalone units of 30 to 140 hours, and the AI leadership units are Level 5. A non-levy employer gets them fully funded up to a modest cap, and they carry no formal qualification. They are short, though: a few dozen hours end before real work throws up its first hard case, which is where most of the learning happens. A unit fits an employer who needs a small, fully funded, one-off push and nothing more. It is not what we would pick when the goal is the real capability a champion needs.

Vendor programmes (Microsoft Copilot adoption, Google Cloud training, and the like) are fine, and often the fastest way to get a team confident in one specific tool. Their limit is that they teach the vendor's product, not your team's own work and data. Use one when you mainly need confidence in a single tool. Use a work-based, tool-agnostic programme when you need people who can frame and solve a problem whichever tool is in front of them, and keep that skill when the tools change.

FAQ

What is the difference between an AI champion and AI awareness training?

An AI champion is a person who does real work with AI and helps colleagues do the same. Awareness training is content that tells people AI exists. The difference matters because awareness alone rarely changes how anyone works: 95% of generative-AI pilots returned no measurable profit or loss, with learning named as the barrier (MIT NANDA, "The GenAI Divide: State of AI in Business 2025"). A champion closes that gap by showing colleagues how, on the team's own work.

How many AI champions does an organisation need?

Fewer than most people expect, and trained deeper than most people plan for. A few people go deep through sustained coached practice; the many around them get working fluency from shorter courses. One well-backed champion per team, supported by a fluent team, does more than a dozen thinly trained ones with no depth to call on. Too few fluent people around a champion and they become a bottleneck who burns out, so the balance matters as much as the number.

How do you choose who becomes an AI champion?

Pick by pull: the person colleagues already go to for help is where a new way of working spreads from, whatever their title. Look for existing trust, real curiosity about the work rather than the tool, and closeness to the regular tasks they can practise on and show. Their job title is a weak guide; where people already turn for help is a strong one.

How long does it take to build practitioner-level AI capability internally?

Real capability takes a programme, not a workshop. Our AI Transformation apprenticeship, a Level 4 route on the IS Business Analyst standard (ST0117), runs about 15 months of training plus a 3-month end-point assessment, judged on real work through a portfolio and a professional discussion. Working fluency for the wider team is much faster, through short courses. The trick is to run both timescales side by side rather than expect depth from an afternoon.

Can we fund an AI champions programme through the apprenticeship levy?

You can fund the depth training through approved data and AI apprenticeship standards using your Growth & Skills Levy account. Training is fully funded for an apprentice aged 16 to 24 at any employer; for a start aged 25 or over at a non-levy employer, the government pays 95% and you pay 5%; and a larger employer can pass up to 50% of its unused levy to cover a smaller employer's cost in full. Skills England maintains the standards, so check the current funding rules there. Funding is only how it is paid for; the reason to run it is capability.

Can we build champions through a vendor programme instead of an apprenticeship?

You can, and for confidence in one specific tool it is often the fastest option: Microsoft Copilot adoption programmes and Google Cloud training, for example, are real routes. Their limit is that they teach the vendor's product rather than your team's own work and data. If you need people who can frame and solve a problem whichever tool they use, a work-based, tool-agnostic programme fits better. If you mainly need a team confident in one product, a vendor course may be all you need.

Why do AI champion initiatives fail?

They fail for three reasons. First, training everyone the same shallow amount, so no one is deep enough to coach anyone. Second, using champions as messengers who forward slide decks instead of showing people how. Third, and most common, naming champions but giving them no protected time, which turns a motivated person into a frustrated one. Give the role real hours, and it has a chance the first two never get.

If you lead transformation, HR or L&D and you are working out how to make AI stick, team by team, start with a conversation: what your teams need to be doing differently in six months, and which routes fit. Our AI Transformation apprenticeship builds the depth a champion needs, and shorter data and AI fluency courses give the rest of the team its footing. Talk to us.