Most AI initiatives in the training provider sector do not stall because the technology fails. They stall because the people side of change is treated as an afterthought. This article sets out why that happens, what the warning signs look like, and how to lead the human side of an AI transition more effectively.

It covers why AI programmes lose momentum, what you are really trying to change, the impact on your team, five common mistakes in AI change leadership, and a practical framework for getting it right.

Why AI initiatives stall

Research from McKinsey suggests that 88% of organisations use AI somewhere, yet very few capture measurable value at scale. Gartner has reported that more than 30% of AI pilot projects are abandoned. The technology is not usually the cause.

A familiar pattern emerges: a small number of enthusiastic individuals adopt a tool, demonstrate some value, and the organisation then tries to scale it to everyone. That is the change leadership jump, and it is where most programmes break down.

AI is a capability. The strategy is what you are changing and how you plan to bring your people with you. Without that distinction, even well-funded AI rollouts produce little return.

Warning signs that you have a change problem, not a technology problem

Several patterns signal that a change management gap is the root issue. If more than two of the following apply in your organisation, the technology choice is unlikely to be at fault.

• AI ownership is concentrated in two or three individuals, with no real adoption beyond them.
• Licences have been purchased but nobody’s role or responsibilities have changed to reflect the new tools.
• Staff are using personal accounts or unofficial tools because nothing formal was offered, or was offered too late. This is known as shadow AI.
• The only AI guidance available is a list of things staff must not do, with no guidance on how to use the tools well.
• One person holds all the knowledge. If they left tomorrow, progress would stop.

What you are actually changing

Before selecting a tool, it is worth being precise about the outcome you are trying to achieve. Name the efficiency gain or the scale you want to drive, not the tool you plan to use. Treat your operation as one system and map the most painful points end to end.

A common mistake is to hand AI the same process you run today. Automating a broken or inefficient process at speed produces more of the same problem. The process itself usually needs to change first, and that requires a deliberate redesign rather than a new tool.

This distinction matters for embedded AI too. Switching on AI capabilities within a platform you already use does not deliver the benefit by itself. The return arrives when you reconsider how the work is done and bring your team along with that changed process.

The impact on your people

Job replacement anxiety is real and reasonable. People are already thinking about it whether or not their employer raises the subject. Avoiding the conversation does not reduce that anxiety; it tends to increase it. Transparency about what is changing and why is more effective than reassurance without substance.

In most providers, the near-term case for AI is productivity, not headcount reduction. The sector does not have a lot of slack. Most staff are already stretched, and there are things that have never been done simply because there has not been enough capacity. AI creates the capacity to address those.

The more concrete you are about this, the easier it is to build genuine buy-in. The same principle applies at a regulatory level. Ofqual has confirmed that AI can support assessment processes, but that summative judgement must retain a human in the loop. Understanding where AI can and cannot operate in your context is part of the honest conversation with your team.

Five things the sector gets wrong about AI change leadership

These mistakes are not unique to training providers, but they are common across the sector.

1. Treating AI as a rollout, not a redesign
Switching from one system to another with similar functionality is different from introducing AI. AI changes how people interact with their work. That requires a redesign of both the process and the way people engage with it.

2. Saying no instead of saying how, and safely
A policy that lists what AI cannot be used for, without explaining what it can be used for, leaves people without a map. Guidance that builds capability explains both sides.

3. Waiting for a mandate and then expecting compliance
Mandating tool use without building awareness and desire first produces compliance at best and resistance at worst. People who have not understood why they are doing something will not do it well.

4. Starting too big
Incremental, measurable changes are far easier to manage and learn from than large transformation projects. Start small, demonstrate real value, use that to build momentum, and keep a longer-term goal clearly in sight.

5. Funding the licences but not the time and training
Buying access to a tool does not mean people know how to use it or have the time to learn. Building in protected time to experiment, alongside training that covers both capability and safety, is the step most organisations skip.

A practical framework: ADKAR

ADKAR is a widely used change management model with direct application here. It breaks the change journey into five stages: Awareness, Desire, Knowledge, Ability, and Reinforcement.

Most AI programmes jump from senior leadership awareness to ability, skipping the desire and knowledge stages entirely. The result is a well-funded rollout with low adoption, because the people affected have not been brought into the why before they were expected to deliver the how.

Desire is the hardest stage and the most commonly skipped. It requires a genuine conversation with the people who will be affected: what is changing, why it is changing, and what is in it for them. That conversation, had clearly and honestly, is the foundation on which knowledge and ability can be built.

The reinforcement stage at the end of the model is also underused. If you are already partway through a rollout and adoption is lower than expected, starting with reinforcement, listening to what has not landed and why, can help you re-run the earlier stages more effectively.

Watch the webinar below to learn more and download the action plan with the ADKAR framework.

Change management action plan