I know I need AI, I don't know where to start. Where do I start?

Start with the problem that already costs you a known number, not with the technology. The best first AI project is the most expensive problem you can measure today. The number becomes the success criterion from week one.

The rule holds when three things are true. The problem hurts in money, shelf-out, loss, queue, margin. You already measure the number, even if poorly. And someone in the operation will use the result, not just receive a report. Without those three, the project becomes a pretty demo with no owner.

The exception is more common than it seems. Sometimes nothing is truly measured, and “I need AI” hides a measurement problem, not a model one. There the first project is not AI. It is instrumenting the number, because most projects fail by starting without it. AI on data nobody measures is an expensive guess.

What you can do this week costs nothing. List the five numbers that keep you up most and mark which you measure today with confidence. The most expensive one you measure is your starting point. The most expensive one you do not measure is project zero. And remember: the first project forms the team that takes over, so it matters more than it seems.

If you are at this moment and want a second read, send me in one sentence the number that hurts most in your operation. I will reply within a business day with a paragraph: start here, or “first, measure this”. No calendar, no call. If a project shape surfaces, we talk.