Most organisations that have been through a major technology transformation will recognise the pattern. The technology gets built, processes are redesigned, pilots are launched and people are trained. There is a lot of activity around the launch.

Then, a few months later, adoption starts to level off.

In our experience, the problem at that point is rarely the technology itself. More often, people don't have a clear enough understanding of what is changing in their own work, managers aren't quite sure how to lead the change in practice, or the organisation is still operating in ways that make the new behaviours difficult.

AI makes this particularly visible. The technology is developing so quickly that traditional approaches to training and change management struggle to keep up.

AI changes the work, not only the tools

One of the mistakes we see is treating AI adoption primarily as a question of giving people access to tools and teaching them how to use them.

That's necessary, but it doesn't get you very far on its own.

The more interesting question is what happens to the work itself. In many roles, people will gradually spend less time producing the first version of something and more time directing the work, evaluating outputs, applying judgment and improving quality. That requires different skills, but also a different mindset.

This is why experimentation matters. People need enough room to try AI in their actual work, figure out where it helps and where it doesn't, and share what they learn with others. A training course can support that process, but it can't replace it.

Not everyone will move at the same speed

In any large organisation, people respond to change differently.

There will be people who are already experimenting before an official programme even starts. There will also be people who are sceptical, sometimes for very good reasons. Most people sit somewhere between the two.

That middle group is particularly important. They usually don't need more excitement about AI. They need to see useful examples, understand what is expected of them and have enough support to start changing their own work.

The early adopters can help here. When colleagues see someone they trust using AI successfully in a familiar context, it is often much more convincing than another presentation about the potential of AI.

Trying to push the entire organisation through the same journey at the same speed rarely works.

The change also looks different depending on where you sit

Senior leaders, people managers and specialists all have a role in AI adoption, but it isn't the same role.

Leadership needs to make the direction clear and, importantly, show through its own behaviour that learning and experimentation are expected. Managers then have to translate that direction into the reality of their teams. They are often the ones answering the practical questions: What should we do differently? What is okay to experiment with? Where do I still need human judgment? What happens if I get it wrong?

For specialists and individual contributors, the challenge becomes much more concrete: finding where AI genuinely improves their own work and gradually rebuilding workflows around that.

This is one reason a single organisation-wide AI training programme is unlikely to be enough. Different groups are solving different problems.

It helps to know where you actually are

We look at AI maturity across six areas: organisation and culture, skills and competencies, operating model, governance and risk, architecture and technology, and AI-specific financial operations.

Across those dimensions, an organisation can be anywhere from Experimenting through Adopting, Integrating and Scaling to what we call AI-native.

The label itself isn't particularly important. What matters is being realistic about the current situation.

An organisation might, for example, have sophisticated technology and governance while still being relatively early in skills or operating-model change. Looking at maturity dimension by dimension makes those gaps visible and helps decide what to focus on next.

It is also worth repeating the assessment over time. The problems you need to solve when a handful of teams are experimenting are very different from the problems that appear when AI starts becoming part of everyday operations.

Measure whether work is actually changing

It's easy to measure activity: how many people attended training, how many AI licences were activated or how many use cases were identified.

Those numbers can be useful, but they don't tell you whether the transformation is working.

We are more interested in questions such as: Do people understand what AI means for their own role? Are they actually changing how they perform specific tasks? Can we see measurable time or quality improvements in those tasks? Are the new practices still in use several months later?

And perhaps most importantly: what happens when the transformation programme is no longer there to push things forward?

If the new ways of working disappear with the programme, the change was never really embedded.

Slow adoption is often a practical problem

When adoption is slower than expected, it's easy to conclude that people are resistant to change. Sometimes they are. But often there is a much more practical explanation.

A team may have learned a new way of working but still be measured according to the old one. A manager may encourage experimentation while people feel that mistakes are punished. Training may be too generic to help with the actual work. Or people may simply have no time to experiment because everything they were doing before the transformation is still on their plate.

These are different problems and they need different responses.

That's why listening mechanisms matter throughout the programme. Some obstacles come up naturally in conversations with teams and managers. Others only surface when people have a genuinely safe or anonymous way to raise them.

What we've learned from change programmes

Leadership behaviour matters more than another leadership message. People managers deserve a lot of attention because they turn an organisation-level ambition into everyday decisions. Training works much better when the organisation also removes the practical barriers to using what people have learned.

We've also found that small examples of real progress can be surprisingly powerful. A team saving a meaningful amount of time on a frustrating weekly task may do more for adoption than a large catalogue of theoretical AI use cases.

And measurement needs to start early. If you only decide what success looks like at the end of the programme, it becomes very difficult to know what actually changed.

Ultimately, the aim shouldn't be to make an organisation dependent on a transformation programme — or on the people supporting it. The aim is to build enough capability inside the organisation that it can continue learning and changing on its own.

If you're working through an AI transformation and want to understand where your organisation currently sits, we're always happy to compare notes and talk through what you're seeing.