The AI Champions Gap: Why You're Training the Wrong People First

Here is what a good AI rollout looks like from the outside: tools procured, licences distributed, a training session booked in. Usage numbers climb. Leadership dashboards show adoption is increasing. Six months later, nothing much has changed.

The problem is rarely the tools. It is the order.

Most organisations start by training everyone. The logic makes sense: if everyone has the same access, adoption should follow. In practice, it does not work that way. Rolling AI capability out uniformly, without building any internal expertise first, is the fastest route to the adoption plateau now affecting the majority of enterprise AI programmes.

A different model is emerging. It is not new in principle, but it is underused in AI contexts. It is the champions model, and the evidence is starting to show that it works.

Most organisations that do invest in a champions model still underinvest in what champions actually need to succeed. Three gaps appear consistently.

The first is permission. Champions are often selected and trained, but not given explicit authority to make decisions about how AI is used in their teams. Without that, they become enthusiasts who cannot change anything. The role needs to come with a mandate, even a narrow one. Someone has to have made a decision that this person's guidance carries weight.

The second is structured peer support. Champions left to figure things out in isolation burn out or disengage. The most effective models pair champions training with a regular forum where champions can share what is working, surface problems, and develop shared guidance. That forum does not need to be heavyweight. A fortnightly call and a shared communication channel is often enough. The value is not the format; it is the signal that this role continues beyond the initial training.

The third is honest technical grounding. Not deep technical training, but enough to be credible. Champions need to understand what AI systems are actually doing, what their limitations are, and how to evaluate an AI output with judgment rather than accept it uncritically. HBR research published this year found that employees simultaneously believe in AI's business value and fear what it means for their identity and sense of competence. Champions who lack technical grounding tend to reinforce that anxiety rather than resolve it. They default to enthusiasm when what teams need is reasoned guidance.

A recent case study from Lebara stands out precisely because it took a different approach. Rather than distributing Copilot to the entire workforce and hoping behaviour would change, Lebara trained a cohort of internal champions and had those champions lead AI adoption across their teams. The difference in outcome was not incidental. It was structural.

This reflects something practitioners working in AI adoption are seeing consistently: the gap between tool deployment and actual behaviour change is wide, and it is not closing by itself.

The numbers back this up. IDC research published this year puts a concrete figure on the global AI skills gap: 5.5 trillion dollars in lost performance. More striking is the internal gap: 77% of workers expect AI to affect their careers within five years, but only 31% have received any meaningful preparation for it. That 46-point gap is not a skills delivery problem. It is a sequencing problem. Organisations are distributing tools before they have built the internal infrastructure to absorb them.

Deloitte's research adds a further dimension: 83% of generative AI pilots fail to reach production. When they stall, it is rarely because the technology did not work. It is because no one inside the organisation was equipped to own adoption, resolve friction, or translate tools into changed workflows. That is the champions gap.

When AI champions appear in corporate communications, they usually mean one of two things: enthusiastic early adopters who happened to find the tools useful, or technically proficient staff selected for extra training because they were already comfortable with technology.

Neither of these is a champions model. They are the absence of one.

A real champions model begins with a deliberate decision about who to develop first, what capability to build in them, and what role they play in the wider organisation. Champions are not heavy users. They are internal translators who can bridge between AI capability and operational reality in their specific context.

That distinction matters because the problems that stall AI adoption are not technical. They are contextual. A champion in a housing association understands which workflows contain sensitive tenant data and why shadow AI is a governance risk in that environment. A champion in a manufacturing business understands safety protocols and why the instructions given to AI tools need to account for quality and compliance requirements. A champion in an HR function understands the emotional dynamics when AI is perceived as a threat to job security.

These are not problems a generic AI training session solves. They require someone inside the organisation who understands both the technology and the specific environment it is being deployed in.

There is a parallel conversation to the adoption question, and it is more urgent than most organisations recognise.

The latest research suggests 78 to 86% of employees are already using unapproved AI tools at work. In housing, the risk is data protection. In manufacturing, it is quality control and safety. In professional services, it is client confidentiality and compliance. In every sector, the workforce has already started without being asked. The question for leaders is whether the organisation helps people do it safely, or finds out what has gone wrong when something does.

Champions are the most effective mechanism for managing this, because they are the people with the credibility and context to have those conversations. A top-down governance policy lands differently when it is communicated by someone who works in the same team, understands the same pressures, and can explain the reasoning in operational terms rather than legal ones.

This is not a secondary benefit of the champions model. For many organisations, it is the primary one.

The organisations seeing sustained AI adoption share a recognisable pattern. They do not start with the whole workforce. They start with a small, carefully selected cohort, trained to depth rather than breadth. That cohort is visible to leadership, adequately resourced, and connected to stakeholders who can remove blockers when they emerge.

Over time, champions become the connective tissue between AI strategy and day-to-day work. They translate policy into practice. They identify where tools are creating friction rather than value. They spot governance risk before it becomes a problem. They build the peer trust that organisation-wide change programmes rarely achieve on their own.

The organisations that get this right are not necessarily the largest or best-funded. They are the ones that asked a different question at the start: not 'how do we get everyone using AI?' but 'who do we need to develop first, and what exactly do they need to be able to do?'

If your organisation has deployed AI tools and adoption is slower than expected, the issue is almost certainly not awareness. People know AI exists. It is visible, discussed, and impossible to ignore.

The question is whether anyone inside the organisation is equipped to make adoption real. Not just to use tools themselves, but to help others use them safely and with judgment, and to identify when AI is being applied in ways that carry risk.

That is what a champions model builds. It is not a training programme you run once. It is an internal capability you develop over time. And the organisations building it now are the ones that will not still be stalled at proof of concept in two years.

If you are thinking about how to structure this for your organisation, we are happy to talk through what a champions programme looks like in practice. The starting point is usually simpler than people expect.

uptakeAI helps organisations build internal AI capability through champions programmes, immersion experiences, and governance design.