The work is in the room, not in the diagram

Why we do this

Real work lives in people's heads. It lives in the quiet decisions an operator makes before lunch, in the email thread that overrides the SOP, in the spreadsheet someone keeps on their desktop because the official system can't hold the shape of what they actually do.

Traditional process interviews flatten all of that. A consultant arrives with a template, asks tidy questions, and walks out with tidy answers, none of which describe the job. The texture goes missing. The exceptions, which are usually where the value and the risk both sit, get rounded off.

AI, used carefully, captures that texture instead of erasing it. Not by replacing the conversation, but by listening alongside it. The operator thinks aloud at the pace of their real work, and the model holds onto the parts a human interviewer would have to summarize away.

What we listen for

The org chart's version of the process isn't what we're after. We're looking for the gap between the official version and the real one. That gap is where the interesting things happen.

• The moments where the documented workflow and the actual workflow diverge.
• The workarounds that have hardened into habit.
• The tribal knowledge, the "you just know after a while" kind, that nobody has written down because nobody was asked the right way.
• The handoffs that look clean on paper and are negotiated by hand every day.
• The reasons people keep the spreadsheet, the side channel, the personal checklist.

These aren't failures of discipline; they're the work itself. Any system worth building has to start by taking them seriously.

What we refuse to do

There is a version of this where AI is pointed at a transcript and told to produce a diagram, and the diagram is then treated as the answer. We don't do that.

• We don't replace the operator's judgment with a model's summary. The operator is the source of truth; the AI is a notetaker that doesn't get tired.
• We don't pretend a clean diagram is the work. A clean diagram is a hypothesis about the work, and it needs to be checked against the people who do it.
• We don't let the AI do the thinking. The synthesis, what matters, what's structural, what's noise, is done by a human expert in the loop, with the model's output as raw material rather than as a verdict.

The technology earns its place by making careful listening cheaper. It doesn't earn it by pretending to listen for us.

What this unlocks downstream

Capturing how the work actually happens is the precondition for everything else we do. In order:

1. Automation. Once the real process is on the table, workarounds and all, the manual grunt work inside it becomes visible and addressable.
2. Integration. Connecting systems only works when you know which handoffs are load-bearing and which are habit. The map tells you where the wires need to go.
3. System replacement. Eventually, when the legacy core is ready to be retired, the replacement has a fighting chance of fitting the work, because the work was understood first.

We lead with listening, not with replacement. The rest follows from there, in that order, and not before.

The service framing of this same work, what it looks like as a thing we deliver, lives on the think-aloud AI capability page.