The scenario is familiar. A demo impresses the executive committee. A pilot is launched, often with enthusiasm. Six months later, it is still running “in test”, used by a handful of people, with no budget to go further. Then it quietly fades away.
The phenomenon is widespread. According to a RAND Corporation report, by some estimates more than 80% of AI projects fail, twice the rate of IT projects that do not involve AI. Gartner, for its part, predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value.
Three causes, almost always the same
The strategy is unclear
The pilot was chosen because it was feasible or spectacular, not because it addressed a business priority. Nobody can say which number it was supposed to move. When the time comes to decide on a larger investment, leadership has nothing to base the decision on. RAND itself puts a misunderstood or miscommunicated problem at the top of the causes of failure.
Deployment is forgotten
A prototype that runs on a laptop, with a hand-cleaned extract of data, is not a system. Going into production takes something else: connection to existing tools, up-to-date data, security, monitoring, maintenance. When none of this has been planned, the cost of the next step surprises everyone.
Teams are left out
The pilot was built for the business teams, but without them. It does not fit the way they work, nobody has been trained, nobody owns it. At the first difficulty, everyone goes back to their old habits.
A pilot that never ships doesn’t count. It has to be designed for production from day one.
Designing for production from day one
In practice, this changes four things as early as the scoping phase.
- A named business owner, who carries the project and answers for its results.
- Written go-live criteria: under what conditions does the pilot become a production system, and under what conditions do we stop it?
- Planned integration with existing tools, data and processes, before writing the first line of code.
- A handover plan: who will maintain the system, with what training and what documentation.
What industrial R&D taught us
Our years of applied research in industry, notably on critical systems, taught us one simple thing: between a lab result and a system used every day, there is work as important as the research itself. Robustness, explainability, integration, adoption. That is the work most pilots skip. It is the work we put at the center.
Will your pilot reach production? Five questions
- Do you know which number it should move, and by how much?
- Is someone on the business team accountable for it?
- Are the go-live criteria written down?
- Is its integration with existing tools planned and budgeted?
- Do you know who will maintain it a year from now?
Fewer than four yes answers: there is still time to reframe.
Why act now
Organizations that are first to move from demo to production gain an advantage that is not only operational: they build skills, data and a culture that their competitors will take years to catch up with. The goal is not to launch more pilots, but to see a few of them through to the end.
- RAND Corporation, “The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed”, J. Ryseff, B. F. De Bruhl, S. J. Newberry, August 2024
- Gartner, “Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025”, press release, July 29, 2024
Let’s sort through your pilots and decide which ones to scale.
Request a diagnosticMore notes