A COO opened a call last spring with three platforms and a pilot already on the calendar. The slide said quick win. We asked to see the work the pilot was supposed to replace. She shared her screen. The work was a spreadsheet, a Slack thread, and a Tuesday meeting where two people argued about which number was real.
No model was going to settle that argument. The argument was the thing worth designing.
we are not there to pick a model
We look at five things, in this order. What the job actually is, in clicks, not on the org chart. Where it stalls. Who is allowed to be wrong. What data exists, as opposed to what the deck says exists. And whether a tool would help the person doing the work, or just move the mess into a chat window.
If we cannot answer those by watching, we are not ready to recommend software. Recommending it anyway is how a company buys a pilot nobody can explain in September.
If we cannot see the work, we will not tell you to automate it.
bring the person who does the job
Not the person who bought the software. Have them finish one real item while we watch, and do not clean it up first. The stall, the side channel, the number nobody trusts — that is the brief.
Sometimes the answer at minute twenty-five is: do not build this. The process is unstable, the data is a rumor, or the opportunity is a meeting that should have been a decision. We would rather say that on the first call than discover it after the contract.