Every engagement we take starts the same way: two weeks inside your workflow before we propose anything. Clients sometimes push back — "we already know where the problem is" — and they're usually half right. The problem is where they say it is. The shape of the problem almost never is.
Here's what those two weeks actually contain.
Days 1–3: watch the work, not the org chart
We sit with the people doing the work. Not managers — operators. An estimator quoting a job, a care manager assembling a prior-auth packet, a capture lead building a compliance matrix. We time real tasks end to end and write down every system they touch.
Two things surface fast. First, the official process and the real process diverged years ago. Second, most of the time isn't spent on judgment — it's spent hunting for the thing judgment needs: the comparable job, the policy paragraph, the prior version.
Days 4–6: trace the data
For each hunting expedition we saw, we trace where the answer actually lives. ERP? Shared drive? A retired employee's export folder? This is where AI projects quietly die, so we do it before proposing one. If the data a system would need doesn't exist or can't be reached, we'd rather find out in week one than in month three.
Days 7–9: draft the boundary
Now we draw a line around the smallest system that removes the most hunting. Deliberately narrow. "Automate quoting" is a bad scope; "draft quotes for repeat-family work from job history, routed to an estimator for review" is a good one — it names the input, the mechanism, and the human checkpoint.
Days 10–12: write acceptance criteria you can hold us to
Every scope ends as a written set of acceptance criteria with numbers attached: coverage, turnaround, accuracy against the baseline we just measured. These go into the build contract. After launch, the system is measured against them monthly under managed operations.
Days 13–14: the honest conversation
We present three things: the ranked plan, the fixed price for the first build, and — when it's the case — the finding that AI doesn't belong in part of the workflow. Roughly a third of scoping phases end with us recommending against automating something the client expected to automate. That recommendation is the product. The build is optional.
If you want to see what a scoping phase would find in your operation, book a consult. Thirty minutes, no deck.