Insights · Use cases
What becomes better?
The wrong first question about AI is “what can it do?” The answer is: an enormous amount, which is no help at all. The right first question is “what becomes better?” Everything useful follows from asking it early and answering it honestly.
Why do capability questions produce bad lists?
Start from what AI can do and you generate the familiar inventory: summarise, draft, classify, search, recommend, automate. All true, none decided. Capability lists grow; they never rank. That is why so many organisations hold a spreadsheet of forty candidate use cases and no conviction about any of them.
Start instead from what should be better and the list ranks itself, because you are comparing outcomes, and outcomes have owners, numbers and consequences.
What does a strong answer sound like?
Weak: “we will automate document review.” Strong: “customers get a decision the same day instead of waiting weeks, and reviewers make more consistent calls with better evidence.” The first names an activity. The second names a change in the world, and you can tell from across the room which one deserves funding.
The discipline is to keep asking “so what changes?” until you hit something a customer, a colleague or the organisation would actually notice. If you never hit it, that is the answer too.
What are the three levels?
- Customer: a faster answer, a clearer communication, fewer repeated questions, less waiting and less doubt.
- Colleague: less searching and re-keying, better decision support, more time on work that deserves a human.
- Business: lower cost to serve, better risk control, quality that holds at volume.
Strong use cases improve at least two of the three. And underneath them sits a fourth level that outlasts the project: capability. What does the organisation know how to do afterwards that it could not do before?
What are the red flags?
A solution chosen before the problem is stated. A value case that is only hours saved. A team fluent about the tool and vague about the decision it improves. Nobody able to say how adoption will be achieved. And the quiet one that predicts failure best: nobody can explain the outcome in a single sentence a senior colleague could repeat. If it cannot be repeated, it will not be funded, and if it is funded anyway, it will not be defended.
Common questions
How many use cases should we run at once?
Fewer than you have listed. A few that pass the test, each with a named owner, beat twenty half-owned experiments.
What if a case helps only one of the three?
Proceed carefully. Two of three is the practical bar.
Who should apply the lens?
Everyone judging ideas. It works because it is shared: prioritisation stops being politics and starts being comparison. Once an idea reaches delivery the same discipline applies to measurement, because the value is downstream.