Measure first, build second. Four stages, and you can stop after any of them.
The fastest way to waste a year on AI is to start with the tool.
We sit with the real work. Not a workshop about the work, the work itself: the reports being assembled, the approvals waiting, the same question answered again. We count hours rather than collect opinions about them.
Every task gets two scores: what it costs you in hours, and how automatable it honestly is today. Plenty of expensive tasks score badly on the second one, and saying so early saves everyone money.
Only the top of the list. We build against your live cases, including the awkward ones your team quietly works around, because those are the ones that break a demo-grade automation.
Runbooks written for the person who will inherit this, not for us. A named owner on your side. Training for the people who touch it daily. If we disappeared, it keeps running.
Some tasks should not be automated. Some processes need fixing before any tool will help. Occasionally the honest answer after the map is that the problem is a staffing or a pricing problem wearing a technology costume.
You will hear that from us, in week two, rather than in month nine.
