Ask your senior estimator to write down how they price a job.
You'll get a four-step list. Maybe six on a good day. It will be a perfectly logical sequence, technically correct, and completely useless to a new hire.
This is the problem we've been circling for the last five articles. The most valuable knowledge in a service business lives in the heads of three or four senior people, and those people can't tell you what they know. They've internalised it so deeply they no longer experience it as knowledge. It's just how things work.
We've called it the tacit knowledge problem in Why Systems Fail. We put a dollar figure on what it costs in The Hidden Cost of Tribal Knowledge. We named the asset that solves it in What is Codified Operational Intelligence™.
This article is about the how. The actual mechanic, the AI-structured interview, that finally makes tacit knowledge extractable at the speed and quality the business needs.
Why traditional interviews fail
Consultants have been "interviewing" senior staff to capture knowledge for thirty years. It almost never works at the quality the business needs. Three reasons why:
1. The expert can't see their own expertise. Anything they've internalised is invisible to them. When asked "how do you price a job?", they describe the explicit steps and skip every judgement call. "And then you check the scope, obviously..." The "obviously" is the part you need. They don't know it's there.
2. The interviewer doesn't know what to ask. An interviewer who hasn't priced a thousand jobs can't catch the gap. They don't know the senior estimator just skipped over the most important decision in the process. The interviewer captures what's said and assumes the rest doesn't exist.
3. The artifact comes out incomplete. What gets written down is a clean, partial process. New hires read it, get to the first ungated decision point, ask the senior estimator, and the cycle resets. The interview produced a document but didn't transfer the knowledge.
The traditional fix is "do more interviews." This compounds the cost without fixing the structural problem. The expert is still the expert. The interviewer is still the bottleneck. Each pass catches a bit more and misses a lot.
What an AI-structured interview actually does
An AI-structured interview is a 60-to-90-minute conversation with a senior person, anchored to a real job they recently worked on, with three things running in parallel:
- A trained Expansive EDGE operator directing the conversation and making judgement calls.
- Live AI transcription capturing every word.
- An AI co-pilot (the same Claude model we use throughout the engagement) analysing the transcript in real time, flagging the moments that need follow-up before the conversation moves on.
The AI's job, specifically, is to do the thing the human interviewer can't do in real time: pattern-match against thousands of similar process descriptions and flag every place a decision was made that wasn't surfaced.
When the estimator says "and then you obviously check the scope," the AI immediately surfaces: "What are you checking the scope for? What's the criterion? Has the criterion ever changed?" The operator reads the prompt, repeats the question in conversational language, the estimator answers, and three minutes of implicit judgement just got made explicit.
Multiply that by every "obviously" and "you'd just" and "of course you would" in a 90-minute conversation. Each one is a piece of tacit knowledge being caught, surfaced, and converted into an explicit decision rule the next hire can use.
A walkthrough: the senior estimator session
Here's what a typical 90-minute session looks like in practice.
| Minute | What's happening |
|---|---|
| 0–10 | Pick a real job from the last 30 days. Surface the source documents (RFP, quote, internal notes). Set the scene. |
| 10–60 | Walk through the job chronologically. "What did you do first? Why? What were you looking for? What would have changed your approach?" AI surfaces follow-up prompts after every described step. |
| 60–75 | Pattern check. AI proposes the implicit decision tree it's extracted. Estimator validates, corrects, adds the edge cases that didn't come up in the example job. |
| 75–90 | Disaster scenarios. "Tell me about a job that went sideways. Why? Where did the warning signs show up? What did you do differently next time?" This is where the highest-value tacit knowledge lives. |
What comes out the other end isn't a transcript. The AI converts the conversation directly into a structured Playbook section: a decision tree, a set of named criteria, an annotated process map with the judgement calls flagged, and a list of edge cases with the responses the senior person uses.
The Expansive EDGE operator reviews, refines, and pressure-tests the output before it's added to the engagement's working artifacts. The senior person gets a one-page summary to validate. Three days later, the same person could read it back to themselves and say "yes, that's how I actually do it."
Two weeks ago, that conversation would have been four sessions across three weeks and produced something 60% as accurate.
The changed economics
The shift this creates is hard to overstate.
| Capture method | Senior staff time | Operator time | Capture quality |
|---|---|---|---|
| "Senior staff writes the SOPs themselves" | ~20 hrs / process | 0 | ~40% |
| Traditional consultant interviews | ~8 hrs / process | ~24 hrs / process | ~60% |
| AI-structured interview | ~1.5 hrs / process | ~4 hrs / process | ~90% |
These numbers are modelled against the patterns we see across engagements. They're not promises. They are, however, an accurate picture of why the work that used to take six months can now be done in six weeks, with higher-quality artifacts at the end.
The senior person's time was always the constraint. The senior person's time is now manageable. That changes which engagements are economical, which knowledge gets captured, and how often the captured knowledge can be refreshed.
The role of the human
It would be easy to read the above and conclude the AI is doing the work. It isn't.
The AI catches what the human interviewer would otherwise miss. It does not direct the conversation, pick which job to walk through, build rapport with the senior person, judge which edge cases matter, or decide which output makes it into the Playbook. Those are all human judgement calls.
What the AI changes is the density of the conversation. A human interviewer working alone might catch one piece of tacit knowledge for every three the senior person reveals. With AI running in parallel, that ratio inverts. The interviewer catches three for every one missed. Over 90 minutes, that's the difference between a useful artifact and a transformative one.
The role of the human is to:
- Choose the right job to anchor the conversation.
- Build the rapport that lets the senior person speak freely.
- Decide which AI prompts to actually surface and which to skip.
- Translate the AI's structured output into language the team will actually use.
- Take responsibility for the final artifact.
The AI extends the operator. It doesn't replace them. We've made this point in the Privacy Policy and the Terms, and we make it again here: every AI-generated artifact is reviewed by a human operator before it reaches a client. The judgement, and the accountability, stays with us.
Where this fits in ControlShift™
AI-structured interviews are the working mechanic of Stage 3: Capture in the ControlShift™ methodology. Stage 1 (Insights) tells us where to look. Stage 2 (Design) tells us what the target operating model should be. Stage 3 (Capture) is where we get what's in people's heads onto the table. Without AI-structured interviewing, Capture used to be the slowest and lowest-quality stage of the engagement. Now it's neither.
If you've ever sat across from a senior employee and thought "I have no way to get what's in their head out before they leave," this is the mechanic that solves it. Not magic. Not replacing the human. Just making the conversation dense enough to be worth the senior person's 90 minutes.