How Do You Run an AI Champions Programme That Works?

TL;DR:

• AI champions are trusted colleagues who show their teams safe, useful ways to use AI. They do not replace ICT, compliance or formal training.
• Choose people others already go to for help, from every team, site and shift. Include at least one sceptic.
• Agree protected time with each champion's manager in writing. We suggest about two hours a week for the first 90 days.
• Give them a one-page charter, training, a monthly huddle and a named person for every kind of problem.
• Measure changed work and safer practice. A count of prompts tells you very little.

Who this guide is for

This guide is for HR leads, ICT managers, L&D officers and managing directors in Kenyan organisations that have started using AI and want it to spread safely. It is written for a company of 50 people as well as for a bank or county government with many branches.

If you are still deciding whether to build skills internally at all, start with our comparison of in-house and external AI training. This post assumes you have decided to try champions and want to know how to run the programme week by week.

What the evidence says

Peer champions have reasonable support as an idea, and rigorous evidence on AI champions specifically is thin.

We found no controlled study of AI champions and no published results from a Kenyan programme. The strongest related evidence comes from healthcare. A 2019 Cochrane review of 24 randomised studies found that local opinion leaders probably improve professionals' compliance with evidence-based practice, with a median absolute improvement in compliance of 10.8%. All the studies were in high-income countries, and the professionals were usually doctors. Results varied, and none reported costs. The review shows that respected peers can change how people work. It says nothing about AI directly.

Company reports are useful for ideas, and they come with limits:

Champions look like a sensible part of a rollout. Be wary of anyone who promises a specific adoption rate because of them. BCG's 2025 survey adds a related point: the share of employees who feel positive about generative AI rises from 15% to 55% with strong leadership support. Champions cannot make up for leaders who are absent.

When not to start yet

A champions programme will struggle if the basics are missing. Wait until these are true, or fix them first:

• There is at least one approved AI tool that staff can reach.
• There is a short written AI usage policy.
• A senior sponsor will read a monthly report and act on it.
• Line managers will release time.
• Someone owns the programme for at least six months.

Our AI readiness assessment can help you check where you stand.

How many champions do you need?

There is no official ratio. We found none from Microsoft or any independent body, and figures on consultants' websites differ widely.

For comparison, Citi's reported 4,000 people work out at about 2% of its staff, and that group covers more than its AI Champions. One consultant's guide, Lead with AI, citing CSO Online, suggests 5% to 10% of your first AI users, with one lead for every 10 to 20 champions.

Our own suggestion, which is judgement and not research, is one champion for every team of about 15 to 25 people. Coverage matters more than the total: every site and every shift needs someone within reach.

How to choose champions

Choose people whose colleagues already go to them for help. The most enthusiastic AI user may not be the right person.

Ask staff who they turn to. Most of the studies in the Cochrane review chose opinion leaders through the sociometric method, which usually means asking colleagues who they go to for advice. A two-question anonymous form can do this:

1. When you are stuck on a spreadsheet, a hard letter or a confusing process, who do you ask?
2. Who in your team explains things well and does not make you feel small?

Then invite volunteers from that list. Citi invited people to volunteer, and we suggest the same.

Get the manager's agreement before you approach anyone. Otherwise the role can turn into a quarrel over time.

Score your shortlist. This is a working tool of ours, and you should adjust it.

Advanced AI skill is missing from that table on purpose, because it can be taught during the programme.

Balance the group. Aim for different departments, branches, ages and genders. Morgan Stanley chose vice presidents, which may suit a large bank. In offices where juniors find it hard to correct seniors, mix levels and let a respected junior lead at least one session.

Include a sceptic. HubSpot picked sceptics on purpose because they point out why something will fail. A thoughtful doubter improves the group, and colleagues are likely to listen when that person says a tool helped.

The champion's job: a one-page charter

Microsoft's adoption site lists ten things a one-page charter should cover. Here they are, filled in with wording you can adapt.

AI Champion charter

Purpose: Help colleagues in [team] use approved AI tools safely on real work.

What you get: Early access to approved tools, training, recognition in your appraisal, and a stronger skill set.

Who you support: [team, site or shift], about [number] people.

Who makes a good champion: Trusted, careful, clear and honest about limits.

Your responsibilities: Show approved uses, share examples that worked, collect questions, and model checking.

What you do not own: IT support, licences, security or legal approvals, formal training, strategy or business results.

Support and escalation: [named contacts from the escalation table].

Who you work with: Your line manager, the programme lead, ICT and the data protection contact.

Development and recognition: Training, a note in your appraisal, and a thank-you from leaders.

Time and participation: [two] hours a week for 90 days, agreed with your manager, and one monthly huddle. Term: six months, renewable.

What champions should and should not do

A clear boundary protects the champion and makes the role credible.

How to train champions

Train them earlier and more deeply than everyone else, and include how to teach. Microsoft's account of its own programme notes that super-users are not professional trainers and need coaching and materials.

BCG found that regular use was sharply higher for people with at least five hours of training and access to in-person training and coaching. We suggest about two days spread over a month, so that champions try things between sessions.

Nobody becomes an expert in a few days, and champions should say that to their teams. Our article on why corporate AI training doesn't stick explains why follow-up matters.

The operating rhythm

Programmes can fade quietly after launch, so put the routine in the calendar on day one.

How much time is right? Published figures vary. Lead with AI suggests 30 to 60 minutes a week. HubSpot's group reportedly met a couple of times a week to report to senior leaders, and one Microsoft champ ran one or two sessions weekly. We suggest about two hours a week for the first 90 days, then about one. Write the time into the person's objectives, because a role added to a full job tends to be the first thing dropped.

Monthly huddle agenda

1. What worked this month: one example from each team (15 minutes)
2. What went wrong, and what was learned (10 minutes)
3. Top questions from the floor (10 minutes)
4. Tool, policy or data changes from ICT (5 minutes)
5. What goes to the sponsor (5 minutes)

Who to call: the escalation table

Champions should never have to guess. Fill in real names and share this table.

The same-day rule has a legal reason. Under section 43 of the Data Protection Act, where there is a real risk of harm from unauthorised access to personal data, the data controller must notify the Data Commissioner within 72 hours of becoming aware of it. That clock is hard to meet if people hide mistakes.

Five hard moments, and what a champion can say

Practise these in training. The wording is a suggestion.

A colleague admits pasting customer details into a personal account. "Thank you for telling me. We need to tell [contact] today. It is much better that you said so now." Then escalate.

A manager asks which staff are using AI badly. "I can share the patterns I am seeing, and I will not give names. People will stop asking me for help if I do."

Someone wants help with a tool that is not approved. "I cannot help with that one for work material. Tell me what you need it for, and I will send the request through the proper route."

A colleague asks if AI will take their job. "I do not know what will happen in every role. I can show you what it does badly and what it still needs you for. Please raise this with [manager or HR] as well."

People treat you as the helpdesk. "That one is for ICT. Here is how to reach them. I am here for how to use AI on your work."

Running the shared prompt library

A prompt library is useful only if someone maintains it. Without an owner it fills with stale and risky items.

Use a simple table with these columns: Task, Prompt, Tool it was tested on, Owner, Data allowed (none, internal or personal), What to check, Date last tested.

Three rules keep it safe:

• No real names, customer data or figures inside saved prompts.
• A champion tests each prompt before it goes in, and a second person checks it.
• Delete anything not tested in the last six months.

Making it work in Kenyan conditions

The basic design travels well, and a few local details need planning.

• Branches and upcountry sites. Give every location its own champion, even if that means more than the ratio suggests. A champion in Nairobi cannot sit beside someone in Kitale.
• Shift workers. Hold the weekly slot during handover, or repeat it for each shift.
• Unreliable connections. Keep offline copies of examples, and consider data support for field champions.
• WhatsApp groups. Staff are likely to share tips there anyway. Agree a rule that no customer data, staff data or confidential documents go there, and keep the official library elsewhere.
• Kiswahili. Encourage demos in the language the team actually works in, and have a fluent colleague check any Kiswahili output.
• Contractors and interns. Include them in the rules, and think twice about giving them the role if they will leave soon.
• Monitoring. Think before you track individuals. The regulator's July 2026 draft AI guidance, as reported by PPC Land, lists employee productivity monitoring among uses that need an impact assessment. Public comment closed on 17 August 2026, and as of 10 November 2026 we found no final version. Our guide to the Data Protection Act and AI has more.

How to recognise and reward champions

Recognition matters, and it can be simple. Citi gives internal badges for completing courses or showing how AI improved a task, and they reportedly carry no promotion or pay rise. Microsoft uses a community badge and public shout-outs.

We suggest a note in the annual appraisal, a thank-you from a senior leader by name, first access to new tools, and a line in the person's development plan. Some organisations may add an allowance. If you do, write the term and end date down so that it does not become a dispute.

How to measure success

Measure whether work changed. Usage counts are easy and tell you little about quality or safety.

Report at team level. Our guide on how to evaluate ROI on AI training shows how to build the numbers into a case for leaders.

What goes wrong, and how to avoid it

A 90-day launch calendar

Five prompts for champions

Remove names and confidential details first.

1. Turn a task into a demo

I am an AI champion in a [department] team in a Kenyan organisation. I have 10 minutes in our team meeting to show how an approved AI tool can help with [task]. Write a short demonstration plan: the problem, a safe example with invented details, the prompt, what to check in the output, and one question for the team. Constraints: no jargon, no real names, and include one thing AI might get wrong. Customisation: colleagues who are nervous about AI.

2. Sort the questions log

Below are questions colleagues asked this month, with names removed. Group them into themes, count each theme, and mark which ones should go to ICT, HR, legal or the champions' group. Constraints: use only what is in the list and do not guess who asked. Customisation: a table I can take to our monthly huddle. Questions: [paste]

3. Check a prompt before it goes in the library

Review the prompt below for our shared library. Check whether it contains any real names, customer data, figures or confidential details. Then suggest placeholder wording, say what a user should check in the output, and suggest which kind of data is safe to use with it. Constraints: do not rewrite it in a way that changes the task. Prompt: [paste]

4. A short note to the sponsor

Turn these notes into a one-page monthly report to our senior sponsor. Headings: What changed in the work, What went wrong, What people are asking, What we need from you. Constraints: use only what is in the notes, no individual names, and no claims of time saved unless a measured figure is included. Customisation: a busy executive will read this in two minutes. Notes: [paste]

5. Prepare for a sceptic

A colleague doubts that AI is useful for [task] because [reason]. Help me prepare for an honest conversation. List where their concern is correct, where AI may still help, and a small safe test we could try together. Constraints: do not oversell and do not invent statistics. Customisation: a respectful conversation between colleagues.

Questions people ask

What is an AI champion?

An AI champion is a trusted colleague who shows their team safe, useful ways to use approved AI tools, collects questions and passes on problems. They keep their normal job. They are not a trainer, an ICT technician or a compliance officer.

How many AI champions does a company need?

There is no official ratio. We suggest about one for every team of 15 to 25 people, and at least one for each site and shift. Start small if you are unsure and grow what works.

Should AI champions come from ICT?

Some can, and most should come from the business. Colleagues are more likely to trust someone who does the same work. Keep ICT close, because champions will need its help and its rules.

Should AI champions be paid extra?

That is your decision, and we found no standard practice in Kenya. Reports say Citi's badges carry no pay rise or promotion. Protected time and proper recognition come first, and any allowance should be written down with an end date.

How long should someone be a champion?

We suggest a six-month term that can be renewed, with some people rotating each round. Rotation spreads skill and limits burnout. It also keeps the programme alive when someone leaves.

Can a small company run an AI champions programme?

Yes, in a lighter form. A company of 30 might have one or two champions, a one-page charter and a monthly conversation with the owner. The time agreement still applies.

Do champions need paid AI tools?

Usually they need only the tools your staff are allowed to use, including free versions where approved. Champions should demonstrate what their colleagues can actually reach. As of November 2026, Microsoft says Copilot Chat without a paid licence can use files you upload, while full access to work data needs one.

How is an AI champions programme different from AI training?

Training gives people skills at a moment in time. Champions keep the practice going afterwards, in the team, on real tasks. Most organisations will need both.

Sources

All sources last checked 10 November 2026.

• Cochrane, "Are local opinion leaders effective in promoting best practice?" (Flodgren et al., 2019)
• AI News, "The quiet work behind Citi's 4,000-person internal AI rollout" (21 January 2026), citing Business Insider
• Bloomberg via Carrier Management, "When Employee AI Adoption Stalls Out, Workplace Influencers Get Called In" (13 November 2025)
• Microsoft Inside Track, "Driving Copilot adoption with our Copilot Champs community" (1 August 2024)
• Microsoft Adoption, "Define your AI champion role and charter", including the Kohler example
• Lead with AI, guide to AI champion programmes (consultant's guide)
• BCG, "AI at Work: Momentum Builds, but Gaps Remain" (26 June 2025)
• The Data Protection Act, No. 24 of 2019 (ODPC)
• PPC Land, report on the ODPC's draft AI guidance note (3 August 2026)
• Digital Policy Alert, timeline of the ODPC consultation on the draft guidance note
• Microsoft Support, Copilot Chat with and without a licence

Where FuKazee fits in

FuKazee trains teams to use AI on the work in front of them, and has trained more than 3,000 people since 2022. We assume nobody in the room knows AI yet, we avoid jargon, and we show the free version of every feature. We never promote one tool, and Microsoft 365 and Copilot are always part of the programme.

For a champions programme, we can train your champions, including how to teach others, and help you set up the charter, the escalation table and the first 90 days. The programme also needs your managers' time and a sponsor who will read the reports, and those have to come from inside your organisation.

See how we work on our page for AI training in Kenya, or our AI training for managers.

To talk about a champions programme for your organisation, email us at lead@fukazee.com