Sector notes
The first use case decides whether the programme survives
MIT's research is explicit that most pilot failures come from poor use case selection rather than model quality. The first two or three carry disproportionate weight, and the instinct about how to choose them is inverted.
Executive summary
6
conditions a first use case should actually satisfy
Core conclusions
- The most consequential decision in an AI programme is which use case goes first — not which model, vendor, or budget.
- The instinct to lead with the most impressive demonstration is close to exactly wrong; MIT's research attributes most pilot failures to poor use-case selection.
- A first pilot should prove AI works in this organisation's specific operational context, not what AI can do in general — ambition belongs in phase two and three.
The most consequential decision in an AI programme is not which model, which vendor, or how much. It is which use case goes first.
The instinct is to choose the one that demonstrates the ceiling — the most impressive thing the technology could do here. That is close to exactly wrong.
What a first use case should actually satisfy
- The friction it removes is real and someone can name it without prompting.
- The data it needs already exists, on the terms production will have, not assembled specially.
- A baseline can be measured before go-live.
- Failure is visible and recoverable rather than silent.
- It touches a process someone owns and wants improved.
- It can be delivered in a window short enough that the sponsor is still in post.
The inverted conclusion
A first pilot should not demonstrate what AI can do. It should prove that AI works in this organisation's specific operational context — which is a different and much more useful claim.
Strategic ambition belongs in the second and third phase, once the capability and the coalition to carry it exist. Programmes that lead with ambition tend to produce an impressive demonstration and no second project.
