Tempo di lettura: 7 minuti
How to interview an expert to capture their knowledge
An expert unconsciously leaves out much of what they know. A capture interview works when it's designed to recover that knowledge, not to ask for a summary.
You ask the person who knows a process best to explain how they do it, and you get an orderly, logical, and surprisingly incomplete version. Not because they're hiding anything. They've automated their own judgment so thoroughly that they no longer see it.
That's why the quality of a knowledge capture is decided in the interview. With the wrong questions, you get a summary you already sensed. With the right ones, what only that person knows, and what never made it into any manual, comes to the surface.
In this article we give you the method to prepare and run that interview: what to do beforehand, which questions get past the expert's blind spot, and how to spot what got left out.
The better you master a task, the less aware you are of the steps you take to carry it out. In studies on surgical training, experts describing a procedure they'd performed hundreds of times left out close to 70% of the steps and nearly three quarters of the decisions.¹ They weren't lying: they had no conscious access to that knowledge.
The consequence is direct. If the question is "tell me how you do it", the answer will skip exactly the valuable part: the judgment, the exceptions, the decision the person makes without thinking about it. The capture interview exists to recover that, and to pull it off you have to design it, not improvise it. It's the most decisive step in the process of capturing internal knowledge and turning it into training.
A good interview starts before the first question. Three preparation decisions:
This is the core of the interview. Instead of asking for a description, you ask in a way that makes the judgment surface. Five types of question that work better than any "walk me through the process":
They anchor the memory in a real episode, where the detail lives.
They surface the mental sequence the person follows without noticing.
They highlight exactly what separates someone who knows from someone starting out.
They capture the unwritten rules, which tend to be the most fragile.
They capture the criteria: when something is right, and when to stop.
That last question is often the most productive of the whole session. It's worth saving for the end, once the person has already gone into detail.
Weak interview. "How do you prepare a new customer's order?" Answer: "I check the details, confirm stock, and send it." Correct, and with nothing that wasn't already in the manual.
Capture interview. "Tell me about the last new-customer order that nearly went wrong." Answer: "I saw the delivery was to an industrial estate, and those carriers won't come up to the plant unless you warn them the day before. I called before confirming." There's the judgment that wasn't written down anywhere.
The difference wasn't the expert, it was the question.
Even when the interview goes well, something is always missing. That's why capture doesn't end when the conversation does.
The first filter is reviewing the recording for jumps: moments where the person says "and then it's done" without explaining how. Each jump is a gap to go back and ask about, this time only on what's missing. AI-assisted capture tools speed up this step: they transcribe the session, order it, and flag where a decision or a reason is missing, so the second round with the expert is short and surgical.
The second filter is validation: have someone else try to carry out the task using only what was captured. Wherever they get stuck, there's knowledge that didn't transfer. It's the fastest way to know whether the capture really worked.
The knowledge that holds up a process is rarely in whoever wrote it down; it's in whoever does it. And that person won't hand it over whole if you ask for a summary, because not even they are fully aware of everything they know.
A well-prepared interview, with questions designed to get past that blind spot, is what turns "I have it in my head" into something another person can learn. The rest of the process (structuring, producing, maintaining) depends on this step being done well.
Between 45 and 90 minutes per task is usually enough. Beyond that, the expert's attention drops and the quality of the material falls. It's better to split a complex process into several short, focused sessions than to force a marathon.
It helps, but it isn't essential. With a prepared set of questions and the discipline of anchoring everything in concrete cases, a team lead or a colleague can run a good interview. What doesn't work is improvising with no script.
It's the most common situation, and it isn't an obstacle, it's the sign that there's valuable tacit knowledge. You solve it by switching register: instead of asking for the explanation, you ask them to do the task while you record, or to tell a specific case. The judgment shows up in the action and the example, not in the theory.
The recording is structured into short, consumable modules and produced in a format the team will actually use. We detail that full path, from capture to a trackable training asset, in the guide to capturing internal knowledge with AI.
