Every new AI engagement starts with the same pressure: someone wants to see something. A demo, a prototype, a proof that the investment is going somewhere. The pressure is understandable. It's also one of the main reasons AI projects stall.
The first 30 days of a well-run engagement are not about output. They're about understanding the actual system — the data that exists versus the data people think exists, the workflow as it's documented versus the workflow as it actually runs, the edge cases that will break any naive implementation in week three.
Week one: understand before you map
The instinct is to start mapping the workflow immediately. Resist it. The first week should be spent with the people who actually do the work — not the managers who describe it, the practitioners who run it every day. They know where the process breaks down. They know which exceptions consume the most time. They know which data fields are reliably populated and which ones are effectively random.
This isn't a stakeholder interview. It's a technical investigation disguised as a conversation.
Week two: map the real system
Now you can map. But map what you observed, not what was described. The gap between the two is usually where the project will live or die.
A good system map at this stage answers four questions: What data exists? Where does it live? What decisions get made against it? What happens when those decisions are wrong?
Weeks three and four: scope the first build
With a real system map, you can scope something specific. Not "an AI solution for intake" — a system that does X with Y data, producing Z output, integrated with W, with these defined success criteria.
The scope document is the deliverable of the first month. Not a prototype. Not a pilot. A document that says: here is the highest- leverage thing to build, here is what it replaces, here is what done looks like, here is what it costs.
Everything after that is execution. The first 30 days determine whether you're executing on the right thing.
[TODO: refine voice — expand with specific examples from discovery sessions; the distinction between documented and actual workflow is the key insight to deepen here]