How to Prioritise AI Use Cases

Companies should prioritise AI use cases on business value at stake, data readiness, implementation complexity, ongoing operating cost and measurable improvement against an existing baseline. Ask whether the data exists today before asking whether the idea is valuable — it is the question that eliminates the most ideas and takes the least time to answer. High-visibility demonstrations should not receive investment priority automatically.

Written by the founders
Drawn from published worked examples

Ask the cheapest question first

Most AI prioritisation exercises start with value. That feels right and it wastes a month. Value is expensive to estimate and cannot be acted on if the data underneath the idea does not exist.

Put every candidate past three questions instead, in this order.

1. Is the data here today?

Not could it be. Is it, right now, in a system you control, with enough history to test against. In the forty-idea example we published, twenty-three failed here. Almost none failed because the data was untidy. They failed because it sat on a client's server, or because the one field that mattered had never been recorded.

2. Are we allowed?

Client consent, regulation, audit trail, personal liability. These conversations take twenty minutes and save twelve-week builds. A statutory certificate issued on the strength of model output is not defensible, and the engineer who signs it is personally liable — that idea took legal eight minutes to close.

3. Does it beat what we already do?

Against the real current cost, including the people. A model does not have to be bad to fail this test. It has to be better than a clerk who is already quite fast, and cheaper than a database query somebody wrote seven years ago and forgot about.

What the survivors need

A stated baseline, a unit cost per task including human review, and a volume. Without those three you have a pilot rather than an investment case.

A model does not have to be bad to fail the cost test. It has to be better than a clerk who is already quite fast.