ADKAR: why 70% of AI projects fail (and how to prevent it)

If you've ever deployed new technology in your company — a CRM, a new management tool, an AI platform — you may know this scenario: three months after launch, the tool is used at 20% of its potential. Your teams have returned to their old habits. The investment was expensive and changed little.

You're not alone. According to Prosci, the global change management reference, 70% of digital transformation projects fail to achieve their objectives. Not because the tech doesn't work, but because no one truly uses it.

With AI, the situation is even more marked. The pace of evolution is fast, promises are strong, adoption is a challenge of its own.

The ADKAR method in two minutes

ADKAR is the acronym of a model Prosci developed to frame the human dimension of change. It describes the 5 stages every person affected by a change must go through to adopt it durably:

1. A — Awareness: the person knows why this change is happening and why now
2. D — Desire: the person wants to participate (personal benefit understood)
3. K — Knowledge: the person understands how to do it in the new system
4. A — Ability: the person can actually do it in practice
5. R — Reinforcement: the new behavior is anchored and sustained over time

Fundamental idea: these 5 stages must happen in order. Skipping a stage guarantees failure.

Why AI projects fail particularly

Here are the 5 classic mistakes I see at SMEs deploying AI:

1. Starting at K, skipping A and D

Natural reflex: ChatGPT Enterprise bought, 2-hour training organized, miracles expected. But without explaining why AI is being adopted (Awareness), or convincing each employee it's good for them personally (Desire), the training won't stick.

Teams listen out of politeness, return to their screen, continue as before.

2. Leadership explains the business why, not the human why

"AI will make us more competitive" — that's the company's why. Necessary but insufficient.

The real why that triggers Desire in your collaborators: "AI will free you from repetitive tasks you hate, so you can focus on what you actually enjoy in your job."

This translation takes 30 minutes of conversation per team. And it changes everything.

3. Training on the tool, not on use cases

"Here's how to use Claude". OK. Concretely, in my job as accountant, what do I do first Monday morning?

Knowledge must be contextualized. Not a generic tutorial on "how to prompt", but: "Here are the 3 prompts you'll use daily for tax monitoring, here's the expected result, here's how you check it's reliable."

4. Confusing Knowledge and Ability

Understanding isn't doing. Your teams can watch the tutorial, nod, and be incapable of reproducing it alone a week later.

Ability is built through practice, individual coaching, real-situation adjustments. Not through a lecture.

5. Forgetting Reinforcement

The project is "delivered": platform running, training done, moving on. Three months later, usage collapses.

Reinforcement means: continuing to measure usage, celebrating first successes, adjusting workflows, working on identified blockers, organizing sharing sessions among users. It's the most neglected phase — and the most critical for durable anchoring.

How I concretely apply ADKAR

On each of my Dubai missions, ADKAR isn't a separate deliverable. It's integrated at every phase:

Phase 1 — Framing (before build): I spend time with the CEO and 2-3 key collaborators. We clarify the business why, translate to human why, identify who'll be enthusiastic, who'll resist, and why. This phase lasts half a day to 2 days based on complexity.

Phase 2 — Build (with progressive involvement): I don't wait for build's end to talk to end users. I show them intermediate prototypes, collect feedback, adjust. By go-live, they've already gotten their hands on it 3-4 times.

Phase 3 — Go-live and contextualized training: No lecture. Short sessions (30-60 min) per role group, focused on the 3-5 most frequent use cases for each role. Immediate practice, no useless theory.

Phase 4 — 30-day Reinforcement: WhatsApp availability for daily questions, short weekly check-in to identify blockers, workflow adjustment if needed, celebration of first measurable time savings.

Concrete result

On projects I've coached with this approach in Europe, 6-month adoption rates exceed 80% — versus an industry average of 30-40% without ADKAR.

In Dubai, where team cultural diversity adds extra complexity (language differences, reference frames, communication styles), this method is even more relevant. ADKAR principles are universal — applying them adapts to each team.

Conclusion

When discussing an AI project, focus is always on technology: which model, which tool, which architecture. Necessary, but that's only 30% of success.

The remaining 70% — human adoption — determines whether your investment truly transforms your company, or becomes a budget line forgotten at next audit.

ADKAR is neither magic nor complex. It's a simple discipline, applied systematically, that radically changes success odds.

If you're launching an AI project, ask yourself: do you have a plan for the 5 ADKAR steps, in order? If the answer is "no" or "we'll see at the end", you now know why 70% of projects fail.