An ExCom calls me in after 18 months and 2.4M€ spent on their "AI strategy". The model runs. Nobody uses it. They want to understand what happened.
What happened has nothing to do with the model. The project was doomed at kick-off, by six decisions made in boardrooms, not in Jupyter notebooks.
Only 21% of AI projects deliver real value (Gartner). Here are the six mistakes I see repeating, month after month, in the remaining 79% — and what I do differently when I arrive early enough to prevent them.
Mistake 1 — Start with tech, not with the problem
"We want to deploy GPT-4." "We want to do generative AI." Nine out of ten briefs land with a tech name. Not a problem.
So you look for a problem that justifies the solution. You always find one — it's just rarely the most important one. A retail group spent 6 months on a customer support chatbot "because everyone's doing it". Their real irritant: 35% team turnover. The chatbot is still there. So is the turnover.
What I do instead: before talking tech, one question in the steering committee. "What problem is costing you the most this year?" Only then do we look at whether AI is the right answer. Often, it isn't.
Mistake 2 — Delegate AI strategy to the CIO alone
AI isn't an IT topic. It's a business + tech + HR + legal + data topic. Delegating it to the CIO alone is like delegating digital strategy to the webmaster in 2005.
The CIO sees technical constraints. Business sees opportunities. Legal sees risks. HR sees human impact. Without that cross-functional view, you get an AI strategy that's actually an infrastructure strategy in disguise — and it crashes against usage the moment it leaves the server room.
The question that unblocks: "who else should be in this room?" If the answer is "nobody, the CIO handles it", you've found the problem. AI governance must be cross-functional — 5 to 8 people, decision-making, meeting every two weeks.
Mistake 3 — Target full automation from the start
"AI will automate the end-to-end process." That's warning sign #2 in my diagnostics. Full automation is a legitimate goal — never a starting point.
The moment you announce "full automation", you trigger: union resistance, team fear, political pressure. The project becomes an HR issue before it becomes a technical one. It dies from resistance long before proving its value.
What I do instead: optimize before transforming. Start in copilot mode — AI recommends, humans decide. Trust builds usage after usage. Eventually, it's the teams themselves who ask for more automation. That's when you shift gears. Not before.
Conceptual shift: don't deploy automation, deploy augmentation. Different verb, different adoption.
Mistake 4 — "We have the data"
Every kickoff where I hear that sentence, I know we're going to lose time. Having data and having usable data is the difference between having a garage and having a car that runs.
I've seen a predictive maintenance project stopped after 4 months and 300K€. The reason: sensors recorded every 5 minutes, the model needed data every 10 seconds. Nobody had verified beforehand.
The test I apply: a 2-week Discovery sprint dedicated to data audit. No model, no code, no dashboard. One question: is the data sufficient, in the right condition, at the right frequency?
That 2-week sprint is the highest-ROI Go/No-Go I know. It typically saves 5 to 10 processes from disaster per audit, and kills the 2-3 dead-ends that would have consumed budget for 6 months.
Mistake 5 — Confuse deployment with adoption
The model is in production. Licenses are distributed. The ExCom announces success. Six months later, nobody uses it. Or worse: teams use it for show but ignore its recommendations.
Deploying a tool and getting it adopted are two radically different competencies. The first is technical. The second is human, managerial, cultural. The second determines ROI — and it's prepared at sprint 1, not after go-live.
What I learned: measure comprehension, not logins. People log in out of obligation. They understand out of choice. The M3K framework structures this around four axes — Mindset, Methods, Metrics, Knowledge.
Mistake 6 — Measure the wrong ROI layer
"The model has 94% accuracy." So what? Model accuracy isn't ROI. It's a metric data scientists love, it says nothing about business value.
Worse: most ExComs steer only the layer 1 of AI value — efficiency. Less time, less cost, more volume. It's measurable, so it's the only one pursued. Two entire layers are left on the table:
→ Layer 1 — Efficiency. Hours saved, cost per transaction, volume processed.
→ Layer 2 — Decision quality. Not how many reports you produce, but how much better-informed, faster, less biased your decisions are. Doesn't fit in a spreadsheet. So nobody looks for it.
→ Layer 3 — Opportunities created. Products you wouldn't have launched, segments you wouldn't have detected, pivots you wouldn't have dared.
Misleading metrics I see in reports:
- "4 hours saved per week" — without measuring what people do with those hours
- "500 documents generated" — without measuring how many are read
- "10,000 tickets auto-classified" — without measuring impact on resolution time
The rule I set at the start of every engagement: every AI project is linked to a business KPI on at least one layer, ideally all three. Did time-to-resolution drop? Did NPS rise? Which strategic decision changed thanks to this model? If you can't make the link, you don't have an AI project — you have a tech project looking for its reason to exist.
Pre-launch diagnostic — the 6 questions
Before investing the first euro in your AI strategy, verify:
- Is the business problem identified and quantified before the tech choice?
- Does governance include business, tech, legal, HR, and data?
- Is the initial scope copilot mode (augmentation), not autopilot (full automation)?
- Is a 2-week data audit planned before any development?
- Does an adoption plan exist from sprint 1?
- Do success KPIs cover at least two of the three value layers (efficiency / decision / opportunity)?
Most organizations I meet start at 1 or 2 out of 6. That's normal. What matters is becoming aware before committing the budget, not after.
To go deeper: IAgile and its 6 principles, and why AI transformations fail because of managers, not technology.
Which of these 6 mistakes did your last AI initiative stumble on?