Experts answer general questions with general answers. Ask “describe your process” and you get the tidy version: the one they would give a new starter, with the interesting parts already smoothed away.
The fix is not better rapport or a longer session. It is asking about specific past instances instead of the general case.
Why does “describe your process” fail?
Because expertise compresses. That compression is what makes someone fast, and it is also what makes them a poor narrator of their own work.
Someone who has run a task a thousand times no longer remembers a thousand occasions. They remember a pattern, and the pattern has the exceptions filed off. When you ask for the process, you get the pattern — accurate, useless, and identical to what is already in the written procedure.
The parts you actually want are stored as episodes. The month the numbers moved. The customer who was an exception. The time the report looked fine and was not. Ask about the episode and the detail comes back with it.
What should you ask instead?
Four kinds of question do most of the work. Each one gets at something the general question cannot reach.
- The last instance. “Walk me through the last time you did this.” A specific occasion produces sequence, timings and names, because the person is recalling rather than summarising.
- The exception. “When would you not do it this way?” Boundaries are where the judgment lives, and people can usually state a boundary faster than a rule.
- The near-miss. “What has gone wrong here, and how did you spot it?” This is the only reliable way to surface the traps, which are invisible during a successful run.
- The reason. “Why is it done this way at all?” The answer either reveals a constraint worth keeping or exposes a rule nobody can justify. Both are useful.
Notice that none of them are about the process. You are collecting events, and the process falls out of them.
How do you follow up?
The first answer is almost never the useful one. The follow-up is the interview.
Four moves cover most situations, and knowing which to reach for matters more than asking more questions.
Probe the hedge. When someone says “usually”, “normally”, “generally” or “in theory”, they have just told you a boundary exists. Ask what makes a case unusual.
Ask for the counter-case. “Has that ever not worked?” turns a confident generalisation into a bounded one, and often produces the best material in the session.
Ask what would change the decision. “What would have to be true for you to do the opposite?” This is the fastest route from an instinct to a rule someone else can apply.
Ask who else would answer differently. People often know exactly where they diverge from a colleague, and will say so if asked directly. They rarely volunteer it.
What do you do with “it depends”?
Treat it as the most valuable sentence anyone will say to you.
“It depends” is not evasion. It is an expert reporting, accurately, that a decision rule exists — and that nobody has ever made them state it. The response is two questions. What does it depend on? And can you give me one case each way?
The first names the variables. The second grounds them in real occasions, so you get thresholds rather than adjectives. Five hundred pounds rather than “a material amount”. Two days rather than “quickly”. A procedure full of adjectives gets applied inconsistently, and that is how two colleagues end up working to different standards without noticing.
How do you keep it on track?
One topic per session, and a deliberate redirect when the conversation wanders.
Experts wander for a good reason: an adjacent story usually is relevant, and it is sometimes more interesting than the question. It still costs you the session. The move is to acknowledge and return. “That’s useful background; coming back to a normal case, what goes wrong?” You keep the material without losing the thread.
Thirty to forty-five minutes on one subject beats two hours across five. Tired people generalise, and generalisations are what you came to avoid. A second short session after you have read the first is worth more than a longer first session, because you will know what to ask.
How many experts should you interview?
At least two per topic, and the second one is where the value is.
One account tells you how one person works. A second tells you which parts are shared practice, which rest on a single person, and where two experienced colleagues quietly disagree. That last category pays for the exercise. Two people applying different thresholds, each certain they follow the same policy, is common — and almost never visible until somebody asks both.
Do not resolve a disagreement by taking the more senior answer. Record both, name who said what, and mark it unresolved until someone with the authority to decide actually decides.
Common mistakes
Five, in rough order of how much they cost.
- Accepting the first answer. It is the summary they have given before. The material arrives in the follow-up.
- Asking about the process instead of instances. Guarantees the pattern, not the detail.
- Interviewing one person. You learn how one person works and mistake it for how the work is done.
- Letting adjectives stand. “Material”, “quickly”, “significant” — each one hides a number the expert could give you.
- Recording without writing anything up. A transcript nobody reads is not knowledge captured; it is the same problem in a longer format.
This is the method whether you run it with a notepad or with knowledge interviews. Those ask the follow-ups and write up the answers, so the second half does not quietly fail to happen. If somebody is leaving, the handover question set is a ready-made place to start.