Why Most AI Transformations Fail — And What a 30-Year-Old Framework Can Teach Us

In 1995, John Kotter analysed over 100 major corporate change efforts at Harvard Business School. Around 70% failed. Not because of bad strategy or wrong technology. They failed because of how organisations tried to lead people through change.

He was writing about ERP rollouts and post-merger restructuring. But when I read his diagnosis today, I see AI adoption described with uncomfortable precision.

McKinsey's 2024 research found that nearly half of AI pilots never successfully scale beyond their initial deployment. Boardrooms are pouring money into platforms and mandating transformation — then watching, confused, as adoption stalls and ROI disappoints.

The technology is not the problem. It never is.

Kotter's 8-step model is not a relic from a pre-digital era. It is a precise map of the human dynamics that determine whether any large-scale change actually succeeds. Here is what each stage looks like in an AI context.

Creating the climate for change

1. Create urgency

Not AI hype. Specific, concrete urgency — which competitors are automating faster, what customer expectations are shifting, what delay will cost in real terms. If your leadership team's sense of urgency comes from a conference keynote rather than a competitive analysis, it will evaporate the moment budgets tighten.

2. Build a guiding coalition

AI owned by the IT department alone will fail. You need operations, legal, HR, finance, and front-line managers — people with credibility across the business who advocate from within. The coalition is not a steering committee that meets monthly. It is a group of people who genuinely believe in the change and have enough organisational influence to make it happen.

3. Form a strategic vision

"Use AI more" is not a vision. People need a human story: what work looks like in three years and how AI serves the organisation's purpose, not just its efficiency targets. A compelling vision answers the question every employee is silently asking: what does this mean for me?

Engaging and enabling the whole organisation

4. Communicate the vision

Kotter found that most organisations undercommunicate change by a factor of ten. The AI vision needs to be repeated, in every format, with senior leaders visibly modelling the behaviours they are asking for. One email and a town hall is not communication. It is an announcement.

5. Remove barriers

Fear of job loss, lack of digital literacy, siloed data, risk-aversion — these are structural blockers. They do not disappear on their own. Leaders have to actively dismantle them. That means investing in training, restructuring data governance, and having honest conversations about how roles will evolve.

6. Generate short-term wins

Not multi-year moonshots. Fast, visible wins that sustain belief and silence sceptics. A team saving two hours a week. An automated report that used to take a day. Small, but real. The purpose of early wins is not ROI — it is proof that the vision is achievable.

7. Sustain acceleration

Early wins create momentum — and complacency. Kotter warned against declaring victory too soon. AI requires continuous reinvestment as the technology itself keeps moving. The organisations that stall here are the ones that treated their first successful pilot as the finish line rather than the starting point.

Embedding the change

8. Anchor it in the culture

This is the most frequently skipped step — and in my view, where virtually every organisation is currently failing.

AI adoption does not become durable until it changes how you hire, promote, and evaluate performance. Until a manager who refuses to engage with AI tools is managed on that refusal. Until AI literacy appears in job specifications as a requirement, not a nice-to-have.

Most leadership teams are not close to this conversation yet. And they are paying for it in stalled adoption rates.

Where most organisations get stuck: In my experience, the majority of companies attempting AI adoption are somewhere between steps 3 and 4. A vision exists somewhere — perhaps in a strategy document or a board presentation — but it has never been genuinely communicated to the people who need to act on it. The gap between having a vision and embedding it across the organisation is where most AI initiatives quietly die.

What it looks like when it works

When BNY Mellon began scaling AI, they did not start with technology. They started with a cross-functional governance board — legal, HR, risk, and operations alongside technology. They ran a narrow, visible pilot in document processing and made the time savings transparent across the organisation within weeks. Adoption followed communication, not the other way around.

That is Kotter's sequence in practice. The coalition came before the tool. The early win was engineered deliberately, not hoped for. The lesson is not that BNY Mellon had better AI. It is that they treated adoption as a change management problem first.

Contrast that with the more common approach: a technology team selects a platform, procurement signs the contract, and an all-staff email goes out announcing "we are now AI-enabled." No coalition. No communicated vision. No urgency that anyone actually feels. Just an announcement.

Kotter would give that initiative a 30% chance of success. Generously.

The competitive advantage is not the AI

The organisations that extract lasting value from AI will not be the ones with the most sophisticated models or the biggest budgets. They will be the ones that can move large groups of human beings through genuine transformation — in sequence, without shortcuts.

That is a high bar. Most organisations are not meeting it. But the ones that do will have an advantage that is genuinely hard to replicate. You cannot licence your way to a change-ready culture.

If you are leading an AI initiative and it feels like it has stalled, the answer is probably not a better platform or a bigger budget. It is an honest assessment of where you are in Kotter's sequence — and whether your leadership team agrees on the answer.