AI TransformationMeasured before you commit
Companies abandon on average 46% of their AI proofs of concept before production, according to S&P Global. We start from your P&L, identify the use cases that can pay off, and treat every pilot as an investment: a price, a stop threshold and an expected return, all set before work begins.
For CEOs who have launched AI pilots or are considering them.
The use cases that can pay off, quantified on your data and ranked by return on investment.
Every pilot has a price, a stop threshold and an expected return, set before work begins.
At your request.
Why so many pilots go nowhere
| Study | Finding | Scope |
|---|---|---|
| MIT NANDA, The GenAI Divide, July 2025 | 95% of organizations see no economic return from generative AI | 300 public initiatives, 52 interviews, 153 survey responses; small sample Source |
| S&P Global Market Intelligence, March 2025 | 42% of companies abandoned most of their AI initiatives, up from 17% a year earlier; 46% of proofs of concept abandoned before production, on average | More than 1,000 respondents, North America and Europe Source |
| IDC, March 2025 | Of 33 proofs of concept launched, 4 reach production | Data, processes and infrastructure not ready Source |
| Gartner, July 2024 | Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. | Data, risk, costs, unclear business value Source |
The causes are always the same: a vague metric, data that isn’t ready, a process no one redesigned, and no one with the authority to call a stop.
Our approach
- Start from the P&L. We find where the value sits, then where AI is the best lever. Sometimes it isn’t, and we tell you so.
- Quantify every use case on your data. Gain, full cost, timeline, risks. We rank them by return on investment.
- Design the pilot for production. Before work starts, we set the metric the company already tracks, the target value, the stop threshold, and the person with authority to change the process.
- Measure two objectives separately. The economic result on one side, team learning on the other. A pilot can train your people. It must not become an excuse for missing results.
- Decide. Scale, adjust or stop. Stopping at the agreed threshold is a successful decision, not a failure. If you want, we then oversee the rollout.
What you can expect
Gains measured on your own metrics, not in demos.
Less money tied up in pilots that go nowhere.
Teams trained by working on their own use cases.
A roadmap your finance team can budget for.
The risks we address at the scoping stage
Incomplete or scattered data.
An unchanged process around a new tool.
Runaway usage costs.
Confidentiality and unsanctioned use.
Compliance with the EU AI Act.
Vendor lock-in.
What a pilot with no results costs
An illustrative example for one company, with its assumptions. Amounts are orders of magnitude.
| Item | Assumption | Amount |
|---|---|---|
| Internal team | 4 people part-time, i.e. 1 full-time equivalent for 4 months, at €6,000 per month fully loaded | €24,000 |
| External provider | 20 days of services, €18,000 in total | €18,000 |
| Licenses, APIs, hosting | 4 months at €1,500 | €6,000 |
| Leadership time | Steering committees | €3,000 |
| Direct cost | €51,000 | |
| Training value to deduct | 15 people, 2 days of training avoided, €21,000 in total | −€21,000 |
| Net direct cost | €30,000 | |
| Opportunity cost | The use case that would have returned €150,000 a year, delayed by 6 months | €75,000 |
| Total economic cost | €105,000 |
The heaviest cost appears in no budget: the deferred gain from the right use case. The learning is real, but it pays back only about 40% of the direct cost. A company that launches five pilots and abandons 46% of them, the average S&P Global found, ties up about €117,000 in direct costs with nothing to show for it.
Frequently asked questions
Why do so many AI pilots fail to deliver?
The studies cite the same causes: data not ready, unclear business value, costs not anticipated, processes not redesigned. A pilot launched with no metric and no stop threshold has no reason to stop, or to succeed.
Isn’t a pilot that trains the team already worth it?
Yes, but rarely in proportion to its cost. In our example, training pays back about 40% of the direct cost. So we set an explicit learning objective, measured separately from the economic result.
Should we start small?
Yes, but on the right problem: a single process, a metric you already track, a quantified target, and a budget for data and training alongside the license budget.
What have your AI pilots returned?
Peter Guillon will call you back within 8 business hours.
