Ship one use case before you write the strategy

The most expensive document in most companies right now is the AI strategy. Not because it cost a lot to produce — though it often did — but because it was written by people who hadn’t yet put a single AI use case into production, and so it’s confidently wrong in ways no one can see until they start building.

We have a rule for engagements: nothing is theoretical past week two. By then, something is running against real data, with real users, doing real work. Everything we believed in week one gets tested, and most of it gets revised.

Why shipping first beats planning first

A use case in production answers questions a strategy deck can only guess at:

1. Where the model is actually reliable. Demos run on the happy path. Production runs on the inputs your customers actually send, which are stranger than anything you’d put in a slide.
2. Where the real cost sits. It’s almost never inference. It’s the integration, the review step, the edge case that needs a human, the data that wasn’t where you thought.
3. Whether your team can carry it. A pattern your team can repeat is worth ten a consultant ships and takes with them.

The shape of a good first use case

Pick something that is high-leverage, low-blast-radius, and measurable.

• High-leverage: it touches a workflow people do all day, so a 30% improvement compounds.
• Low-blast-radius: a mistake is recoverable and visible, not silent and catastrophic.
• Measurable: you can state, before you start, the number that will move.

If you can’t name the number, you’ve picked the wrong use case. Pick again.

Once one thing is shipped, the strategy almost writes itself — because now it’s grounded in something you’ve actually seen work, and the second use case takes a third of the time. That’s the pattern. The strategy is the output of shipping, not the input to it.