The SME Interview: Getting the Judgment Out of the Expert's Head Without Wasting Her Week

Ask your best underwriter, technician, or closer to "help build the training" and one of two bad things happens. Either she produces a slide deck — usually a heroic, unusable tour of everything she knows, organized the way an expert's mind is organized, which is to say uselessly for a beginner. Or she sits for a meandering hour of "so, tell me about the process" and gives you the official version: the same sanitized sequence that's already in the documentation nobody learns from. Both failures have the same root. Expertise is largely invisible to its owner. The checks she runs before acting, the early-warning smells, the four options she weighed and discarded in two seconds — that's the actual skill, and she no longer notices herself doing any of it. Decades of research on expert knowledge says this isn't modesty; it's the structure of mastery. The good news: the invisible parts come back out under the right questioning. That's what the SME interview is for, and done properly it takes ninety minutes, not a week of workshops.

Before the interview: pick cases, not topics

The single highest-leverage preparation is changing the unit of conversation from topics to recent, specific cases. Ask an expert about "handling escalations" and you get philosophy. Ask about the escalation she handled Tuesday and you get the truth — memory for specific episodes carries the judgment that generalizations launder out. So before the session, ask her to jot down (two minutes, no prep beyond this) the last three or four real instances of the task you're training: one that went smoothly, one that got hairy, and if possible one where she intervened before it went wrong. Those cases are the interview's entire agenda. Book ninety minutes, recorder on — transcription is your note-taker so your attention isn't — and tell her the honest framing that makes experts generous: "You're not designing the course. I'm mining you. Just talk about the work."

The protocol: walk the case, then dig where it bends

For each case, one opening move: "Walk me through it — from the moment it landed on you to the moment it was done." Then let her run, and resist the beginner-interviewer urge to fill silences; the pauses are where she's retrieving the parts she never says aloud. Your job is to notice the bends — every point where the story includes a decision, a check, a hesitation, or a deviation from the official process — and dig at each one with the five questions that surface judgment:

1. "What did you look at before you did that?" — recovers the pre-checks and the information she consults without noticing she consults it.
2. "What would have made you do it differently?" — recovers the branching logic; this is the question that turns one anecdote into a decision rule.
3. "How did you know that early?" — recovers the smells: the weak signals experts read that beginners walk straight past.
4. "What would a smart new person get wrong here?" — the single best question in the kit; experts who can't articulate their own skill can describe beginners' failures with forensic precision, and every answer is a training objective wearing casual clothes.
5. "What's the version of this the documentation doesn't tell you?" — asked with a grin, near the end, once trust is up. The answers are why you came.

When she says "it depends"

"It depends" is not a dead end — it's the door to the best material in the session, because dependency structure is expertise. The follow-up is always the same: "Depends on what? Give me a time it went each way." Two contrasting episodes turn the shrug into a rule with conditions attached, which is exactly the form training can teach and a scenario check can assess. The related trap is the expert who answers in shoulds — "you should always confirm the account first" — when her own case shows she didn't. Don't call it a contradiction; call it interesting: "In Tuesday's case you skipped that — what told you it was safe to?" The gap between the official rule and the expert deviation is never hypocrisy. It's the judgment layer, caught in the act.

Close the loop: the teach-back and the wince test

End the ninety minutes with five minutes of teach-back: summarize the three or four decision rules you think you heard — "so the real logic is: check X first; if Y, slow down; Z is the early warning" — and let her correct you. The corrections are gold; experts who can't generate their model will edit yours ruthlessly and precisely. Then you're done with her calendar. What leaves the room with you: a transcript full of episodes, a set of conditional rules, a beginner-mistake list, and quotes with the texture real training needs. When the drafted modules come back — whether you wrote them or AI drafted them from the transcript — her remaining involvement is one pass of the wince test: read it and mark where it oversimplifies. Total expert time: about two hours, for a course that actually contains her expertise instead of gesturing at it.

Where the tool fits

The interview is irreplaceably human — the noticing of bends, the follow-up at the right moment, the trust that unlocks question five. What comes after is not: LearningByDesign takes the transcript and the objectives and produces the structured first draft — modules built around the decision rules, scenario checks generated from the it-depends branches, the beginner-mistake list turned into the judgment layer of each module — so the expensive human minutes get spent where they're irreplaceable and the drafting happens at machine speed. You run the wince pass; the expert runs hers once. Nobody builds slides at midnight.

The bottom line

The expert was never the obstacle — the request was. "Help build the training" asks her to do your job badly; "walk me through Tuesday's case" asks her to do her job out loud, which she'll do happily and well. Cases before topics, bends before steps, five questions at every decision point, "depends on what?" every time, and a teach-back before you leave. Ninety minutes of that beats a month of shadowing and a folder of official process docs — because it captures the one thing those never contain: what she actually checks, actually fears, and actually does when the case stops being typical.

— Tom