The map matters before the machine.
A lot of automation conversations start with the machine too early.
Someone knows the work is heavy. They know the team is stretched. They know there is software involved. So the question becomes what can AI do here, which sounds practical and usually skips the most useful part.
The map is where the truth starts showing up.
When you map the work, the problem often changes shape. The task that looked slow turns out to be waiting on approval. The handoff that looked manual turns out to be protecting judgment. The form that looked broken turns out to be asking the wrong question.
That matters because every system inherits the assumptions underneath it. If the map is wrong, the build becomes confident in the wrong direction.
This is why I keep coming back to conversation as a real part of the work, not a prelude to the work. A good conversation can expose the route, the constraint, the risk, and the place where a person should stay in control.
Sometimes that conversation leads to software. Sometimes it leads to a process change. Sometimes it leads to the uncomfortable realization that the business does not need more automation yet. It needs a clearer operating memory.
That is not a worse outcome. It is the point. The system should serve the work that is actually there, not the version that sounded better before anyone drew the map.
Before you ask what AI should build, can you show where the work actually goes?