Guide · Your team and AI
AI champions: how to get employees to use AI well
AI champions are colleagues who use AI well in their own work and help their team do the same. They need protected time, support and a route to raise problems, not just a badge. Champion networks are widely recommended, but no rigorous evaluation yet shows that they work, so treat them as one part of a wider plan.
By Andrew Ward, Founder of ScorchsoftChecked 11 minute read
In short
- People start using AI well when they have an approved tool and clear rules, time to learn on real work, and colleagues and leaders who use it openly. An order from the top, on its own, is likely to raise usage more than value.
- Give each champion protected time, someone to take problems to and a short written role. A badge on its own changes little.
- A good champion shows one real win and writes it down for everyone. One who answers every question personally becomes a bottleneck.
- Run a ten-minute show-and-tell: one win, one mistake caught, one thing to share and one place you copy and paste. Thank the person who shares the mistake.
- Don’t rank people by how much AI they use. You’re likely to get more usage, not better work.
Written for: Team leaders, managers and owners in UK organisations who want their people to use AI well, safely and in the open.
What is an AI champion?
An AI champion is a colleague, usually not in IT, who uses AI well in their own job and helps the people around them do the same. They show what works on real tasks, help colleagues get started and pass problems to whoever can fix them.
You may also see “AI ambassador”, “digital champion” or “super-user”. This guide uses AI champion throughout. The role is part-time and sits alongside the person’s normal job. It works best inside one team or department, where the champion knows the work well enough to spot which tasks AI can genuinely help with.
| An AI champion is… | An AI champion isn’t… |
|---|---|
| A colleague who shows one real win on the team’s own work | A help desk for every AI question |
| Someone who writes good prompts down once so everyone can use them | The keeper of a private stash of prompts |
| A route for problems and ideas to reach whoever can act on them | The person who approves tools or enforces the rules |
| A volunteer with time set aside for it | An unpaid extra job squeezed in around the day |
How do you get employees to use AI well?
Give people the means, the time and a reason: an approved tool with clear rules, protected time to try it on their own work, and leaders and colleagues who use it openly and share what works. An order to use AI, without those, is likely to raise usage more than it improves the work.
A widely used model of behaviour change, COM-B (Michie, van Stralen and West, 2011), says people do something new when they have the capability, the opportunity and the motivation. It’s a useful checklist for AI at work, because many rollouts supply the tool and little else.
| People need | What it looks like with AI | Who makes it happen |
|---|---|---|
| Capability: knowing how | Basic training on how AI works, how to brief it and how to check its work, practised on real tasks | Manager, with a course or an in-house session |
| Opportunity: the means and the time | An approved tool on an organisation plan, the tramlines (what data may go in) and time set aside to learn | Owner or IT for the tool; the manager for the time |
| Motivation: a reason to bother | Seeing a colleague save real time on a familiar task, and leaders using it openly, mistakes included | Champions and leaders |
The gaps are real. In the CIPD’s Good Work Index 2026, only 13% of UK employees said they had received any training in using AI in their current job (UK employees, from a survey of 5,074 working adults, December 2025 to February 2026). In BCG’s international survey, 55% of employees felt positive about generative AI where leaders strongly supported it, against 15% where they didn’t (more than 10,600 leaders, managers and frontline white-collar employees in 11 countries and regions, published June 2025; consultancy research, and a correlation, not proof of cause).
Gareth manages a 40-person accountancy practice. Under pressure to have “an AI plan”, he tells everyone they must use AI from Monday. Three weeks later, usage is up and the work is no better: people paste in tasks to be seen using it, and Darren, his sceptical payroll senior, quietly stops. Gareth starts again. He gives everyone an hour a week to try the approved tool on their own tasks, asks Hayley, who has been using AI well for months, to show one real win at the weekly meeting, and starts sharing his own attempts, including the ones that went wrong.
Do AI champion networks actually work?
Nobody knows for certain. Champion networks are widely recommended, including in vendors’ adoption guides, but we found no rigorous evaluation of their outcomes. The evidence for what sits around them, peer learning, leaders’ openness and psychological safety, is stronger.
Microsoft’s own adoption guidance recommends AI champion programmes and suggests each one has a charter covering time commitment and “support and escalation paths” (read 29 October 2026). Its outcome story is a single customer case, in which champions were one part of a wider change programme. That is practitioner advice, and vendor advice at that, not an evaluation.
The research on what helps people speak up and learn at work is more solid, and it points the same way. Psychological safety, a shared belief that it is safe to take interpersonal risks such as admitting a mistake, was linked to more learning behaviour in a study of 51 work teams (Edmondson, 1999). Managers’ openness to ideas was more consistently related to employees raising improvements than transformational leadership was, in a study of a US restaurant chain (Detert and Burris, 2007). And one common reason people stay silent is the feeling that speaking up won’t change anything (Milliken, Morrison and Hewlin, 2003; 40 interviews).
What this means in practice
Treat champions as a reasonable bet, not a proven fix. They are more likely to help where leaders make it safe to share mistakes and act visibly on what champions raise. Set yourself a date, say 90 days, to ask whether anything changed.
What does a good AI champion do?
A good champion shows a real win on the team’s own work, writes it down so anyone can repeat it, helps colleagues try it and passes problems on. They improve how the team works, rather than doing everyone’s AI tasks for them.
- Shows one real win at a time. A five-minute demonstration on a task the team does every week beats a slide of possibilities.
- Writes it down once. A good prompt that lives in one person’s chat history is lost when they go on leave. Saved as a shared instruction set or AI skill, with an owner and a review date, it works for everyone. Module 6 teaches how.
- Sits beside colleagues while they try. The aim is that each person can do it without the champion next time.
- Collects the problems. Where the tool falls short, where the rules are unclear, where people still copy and paste between systems. Then passes them to whoever can fix them.
- Says what AI did and what they checked. Champions set the habit of open, careful use.
Watch for the “router” champion
A champion who answers every question and runs every prompt personally becomes a bottleneck, and the team learns nothing. If your champion’s queue is growing, help them turn the common requests into shared instructions and short how-tos, so the team can carry on without them.
What do AI champions need from their organisation?
Champions need protected time, support from a named manager and a route to escalate problems, written down in a short role. Without these, the role becomes an unpaid extra job and quietly lapses.
“Champions need time, support and an escalation route, not just a badge” is how Andrew Ward, founder of Scorchsoft, puts it to clients: his practice from rolling AI out, not research. A one-page role makes it concrete.
A one-page AI champion role
- The team or area they cover.
- Protected time each week or month, agreed with their line manager and shown in their objectives.
- What they do: show wins, write shared instructions, help colleagues try, collect problems.
- What they don’t do: approve tools, police colleagues or do other people’s work for them.
- Who they take problems to: a named person for tools and data, and one for everything else.
- How they share: a slot at an existing team meeting, and a place for shared instructions.
- A review date, about 90 days out, to ask what changed and whether to carry on.
Champions only make sense once people have an approved tool and clear rules. If staff are still using their own accounts quietly, start with shadow AI, then the AI policy template.
How do you run a ten-minute AI show-and-tell?
Add ten minutes to a meeting you already hold. Four people each take a turn: one win, one mistake caught, one prompt or instruction to share, and one place where someone still copies and pastes between systems.
Step 1: One win (3 minutes)
A task AI helped with this week, what it saved and what the person checked.
Step 2: One mistake caught (2 minutes)
Something AI got wrong and how it was spotted. Thank the person. This is the most useful slot on the agenda.
Step 3: One recipe to share (3 minutes)
A prompt or instruction set someone can reuse, saved where the team can find it, with an owner.
Step 4: One glue point (2 minutes)
A place where someone still moves information between systems by hand: glue work that might be worth fixing later.
Run it as Module 7 of The AI Stairway™ teaches: flag mistakes early, blameless about people and strict about the system. When someone shares a problem or an idea, act on it where people can see: try it, explain a no, or give it a route to a test. People stop speaking up when they think nothing will change.
What to say at the first one
“We’re trying ten minutes a week to share how we use AI. I’ll go first, including something it got wrong for me. Nobody has to share, nobody will be judged for a mistake, and if you raise a problem we’ll tell you what we did about it.”
How do you choose AI champions?
Ask for volunteers who already use AI carefully in their own work and are respected by their colleagues. Aim for one per team, and include at least one thoughtful sceptic.
- Ask, don’t appoint. A reluctant champion helps nobody. Offer the role and the time that comes with it.
- Pick for care, not enthusiasm. The best champions check AI’s work and say when it’s wrong. A sceptic who has found one genuine use can be more persuasive than an enthusiast.
- Spread them across teams. A champion who knows the work can see which tasks AI can help with.
- Look for the quiet expert. Someone may already be using AI well and saying nothing. The shadow AI week often finds them.
In a firm of 15 people, you don’t need a network. One champion, with an hour a week and the owner as the person they take problems to, plus the ten-minute show-and-tell at an existing team meeting, does the same job. The owner or manager keeps the shared place where recipes are written down, so the know-how stays when people move on.
In the AI Stairway™ to Success, Scorchsoft’s own model of how organisations adopt AI rather than an industry standard, local champions are one of the moves from step 2, “Informal professional use”, where know-how is trapped in individuals, to step 3, “Shared best practice”, where the best instructions are written once and used by the whole team.
What should you avoid with AI champions?
Avoid usage league tables, champions without time, champions as enforcers, and treating the network as the whole plan. Each of these turns a helpful role into a box-ticking one.
- Ranking people by usage. It’s likely to reward looking busy with AI rather than doing better work. Judge by finished work and what people checked.
- A badge with no time. The role tends to fade once the day job takes over.
- Champions as enforcers. If champions report colleagues’ unapproved use, people stop asking them for help. Rules are a manager’s job.
- A network instead of a plan. Champions can’t make up for no approved tool, no rules or leaders who never use AI themselves. Decide who owns those with AI governance.
- Judging people by course data. Training scores, feelings and how much AI someone uses aren’t measures of their work.
Questions people ask
How much time should an AI champion get?
There is no tested figure. Agree a fixed amount with the champion’s line manager, write it into their objectives, and review after about 90 days whether it was enough. What matters most is that the time is protected, not squeezed in.
Should AI champions be paid extra?
That is your organisation’s call. At the least, recognise the role in objectives, reviews and development conversations, so it counts as real work rather than a favour.
Does every team need an AI champion?
No. Small organisations may do well with one or two, plus a weekly show-and-tell. What every team needs is an approved tool, clear rules and time to learn.
What if staff refuse to use AI?
Ask why. Common reasons are not trusting the output, worry about jobs, no time to learn, or a task AI doesn’t help with. Each has a different answer, and some are right. Forcing use is likely to produce more usage than value.
Can a manager be the AI champion?
A manager can and should use AI openly, but a peer champion can work better: people find it easier to admit what they don’t know to a colleague than to the person who reviews their work.
Training for the whole team
The AI Stairway™ gives everyone the same careful habits, then helps your team agree its rules and test one AI idea properly. Everyone can start with the free introductory module.
AI training for teamsSources
Every figure on this page comes from one of these. Links open in a new tab.
- CIPD Good Work Index 2026 (section 5, generative AI) (opens in a new tab). CIPD, July 2026.
- AI at Work 2025: Momentum Builds, but Gaps Remain (opens in a new tab). Boston Consulting Group (consultancy research), 26 June 2025.
- Define your AI Champion role and charter (opens in a new tab). Microsoft Adoption (vendor guidance), no date on page; read 29 October 2026.
- The behaviour change wheel: a new method for characterising and designing behaviour change interventions (Michie, van Stralen and West), Implementation Science 6, 42 (opens in a new tab). BioMed Central, 23 April 2011.
- Psychological safety and learning behavior in work teams (Edmondson), Administrative Science Quarterly 44(2) (opens in a new tab). SAGE, June 1999.
- Leadership behavior and employee voice: is the door really open? (Detert and Burris), Academy of Management Journal 50(4) (opens in a new tab). Academy of Management, August 2007.
- An exploratory study of employee silence: issues that employees don’t communicate upward and why (Milliken, Morrison and Hewlin), Journal of Management Studies 40(6) (opens in a new tab). Wiley, 2003.