Ask for production numbers, all-in cost, governance, error visibility, and exit terms before anyone falls in love with the demo.
Drawn from AI That Pays for Itself, the book's operator-first method for making AI produce measurable value.
Cost, latency, and accuracy at your volume, on your messy data.
What happens when it gets it wrong, who sees it, and where the exception goes.
Data export, portability, and contract-out terms. You are signing a lease; make sure you can break it.
It goes into Kit so follow-up can match what you actually asked for, not some generic newsletter fog.
Use this before a vendor meeting, before a pilot, or before asking a board for budget. The point is to force the operational questions first: what moves, who owns it, what it costs, when we stop, and how we keep watching after launch.