AI that earns its keep: a checklist for the UAE founder.

Every brand has an AI line item this year. Most will quietly disappear next year. The difference between AI that earns its keep and AI that becomes a sunk cost is usually decided before a single line of code is written.

Start with the smallest possible problem with the largest measurable outcome. If you cannot describe — in one sentence — the hour saved, the error avoided or the revenue unlocked, you do not have a use case. You have a hope.

Every AI feature needs a number it owns. Pick two: one operational, one experiential. The operational number is the dial you would defend to a CFO; the experiential number is the one your customer would notice if it slipped.

Write the evaluation plan before you build. Decide what 'better than today' looks like, who owns the dashboard and how often you'll review it. AI features without measurement are theatre.

What is the single sentence outcome? Who owns it? What data does the model see, and where does that data live today? What is the worst answer the model could give, and what stops it leaving the building? Who reviews edge cases in the first 90 days? How will we measure it? Where does it integrate? What is the rollback plan? What does a five-year operating cost look like? And — honestly — would we still build this if it were a person doing the job?

The answers tell you more than any vendor demo. The shape of the answers tells you whether the team is ready.

Most failed AI initiatives failed on ambition, not technology. Start with the smallest production-grade slice and let it run live for a quarter. Instrument it. Compare it to the human baseline. Then, and only then, scale.