Executive summary
6
conditions a first use case should meet
Key conclusions
- The most important decision in an AI programme is which use case goes first.
- Leading with the most impressive demonstration is almost exactly the wrong choice. MIT's research attributes most pilot failures to poor use-case selection.
- A first pilot should prove that AI works in this organisation's own operations. Ambitious projects belong in phases two and three.
The most important decision in an AI programme is which use case goes first. The choice of model, vendor and budget all follow from it.
The instinct is to choose the use case that shows the most impressive thing the technology could do here. That choice is almost exactly wrong.
A first use case should meet six conditions
- It removes real friction, which someone can name without being prompted.
- The data it needs already exists, in the form and under the conditions that production will use.
- A baseline can be measured before go-live.
- Failures are visible and can be recovered from.
- It changes a process that someone owns and wants improved.
- It can be delivered quickly enough that the sponsor is still in their role at the end.
A first pilot should prove AI works in your own operations
A first pilot should prove that AI works in this organisation's specific operations. That claim is much more useful than a demonstration of what AI can do in general.
Ambitious projects belong in the second and third phases, once the organisation has the capability and the internal support to carry them. Programmes that start with ambition tend to produce an impressive demonstration and no second project.
