The Tech Who Can Do It But Cannot Explain It

Why this matters

Your best diagnostician walks up to a piece of equipment, listens for ten seconds, and names the fault. Ask how, and you get "you just know." That answer is not evasion and it is not modesty. It is an accurate report of what it feels like from the inside, and it is a serious business risk, because the knowledge that makes them your best tech is the knowledge you cannot copy, cannot train against, and cannot keep when they retire or leave. Treating it as a personality trait guarantees you lose it. Treating it as a capture problem means you can get most of it out.

What is actually happening inside expertise

Skill consolidates. Early on, a task is a long conscious sequence of separate decisions. With enough repetitions those decisions fuse into single automatic moves, and the intermediate steps stop being consciously available. The expert genuinely cannot see the steps anymore, the same way you cannot narrate how you catch a thrown object.

Two consequences follow, and both are misread constantly:

• The expert underestimates their own process. They will describe a five-step read as one step, and they are not being lazy. The other four are invisible to them.
• The expert overestimates what a beginner sees. They will say "obviously it was the intake" without registering that the apprentice standing beside them heard nothing unusual at all. This is not arrogance. It is that expertise silently filters the world, and the filter is invisible to the person wearing it.

So the question is never "why won't you explain it." It is "what question would make the missing steps visible again."

Telling can't-explain apart from won't-explain

Different problems, opposite responses, and confusing them is expensive. Signatures:

The extraction methods in this card address the left column. The right column is a management conversation about whether being the single point of failure is an acceptable position in your shop, and it will not be solved by any technique. See related: The Single Point of Failure Audit for Shop Skills.

Why sending them to a teaching class usually misses

The instinct is to fix the expert's teaching skill. That helps a little, but it targets the wrong constraint. The constraint is not delivery, it is retrieval: the content is not accessible to the expert in the first place, so improving how they present it improves the presentation of a gap.

What works instead is putting them back into a situation where the automatic process has to slow down and become conscious again. Every method below does that in a different way, and none of them require the expert to be a good teacher.

Five extraction methods, cheapest first

1. Narrate while doing, with a rule against results. The expert works and talks, and the rule is that they may only describe what they are perceiving and considering, never the conclusion. "The pitch drops when it loads" is allowed. "It's the bearing" is not. Removing the conclusion forces the perceptual steps into words. Costs nothing but a slightly slower job.

2. Pause and predict. The apprentice, not the expert, drives. At each decision point the apprentice states what they would check next and why, then the expert says only "yes" or "not yet, check something else first." When the apprentice is wrong, the expert has to say why, and the why is the buried step. This one gets the most content out per hour, because it targets exactly the places the two mental models diverge.

3. The contrast interview. Ask the expert about the last time they were wrong, or the two cases that look identical and are not. Experts cannot enumerate their rules but they can almost always tell you the exception, and the exception implies the rule. "What would make you not think it was the intake?" gets a precise answer where "how did you know it was the intake?" gets a shrug.

4. Error catalog. Sit for one hour and ask for the five mistakes they see newer techs make on their strongest job type, in order of how often. Every entry names a decision point the expert handles automatically. This is the fastest way to produce a usable checklist for something nobody has ever written down.

5. The scribe pattern. The apprentice writes the procedure, not the expert. The apprentice drafts what they think the steps are, the expert reads it and corrects it, and corrections are far easier to produce than a blank page. Experts who cannot write a procedure can reliably tell you what is wrong with one.

Rank by how much time you have: methods 1 and 4 cost the least and produce the most immediately usable material. Method 2 costs the most job time and produces the deepest content.

What to do with what you capture, and where to stop

Capture only what is rare and consequential. Do not try to document everything the expert knows, because most of it is available in any reference and the effort will collapse under its own weight inside a month.

Three filters, all of which must be true:

• Only this person can do it, or does it noticeably better.
• It shows up often enough to matter, or rarely enough that nobody else has learned it and it hurts when it appears.
• Getting it wrong is expensive: a callback, a safety event, a warranty denial, a repeat visit.

Whatever survives those three filters goes into the shop's own material as a one-page card with the failure mode named. Everything else is fine as an oral tradition.

Worked example: capturing a 20-year tech's intake read

A shop's most senior tech diagnoses a recurring equipment fault correctly at roughly twice the rate of the next-best tech, and about three times faster. Asked how, they say they can hear it. The owner runs four sessions across five weeks, using one method each and one hour each, so four hours total.

Session 1, narrate while doing, results banned. Over two service calls the tech produces eleven perceptual cues that had never been said out loud, including three that turned out to be the load: a change in pitch under load rather than at rest, a specific ordering of the checks, and one thing they smell that they had never mentioned because it did not seem worth mentioning.

Session 2, pause and predict, apprentice driving. Six divergence points in one call. At four of the six the apprentice's next check was reasonable but out of order, and the expert's correction each time was a version of the same rule: check the cheap reversible thing before the thing that requires teardown. That rule had never been stated in the shop and it was the single most transferable item of the whole exercise.

Session 3, contrast interview, one hour at the shop. "What would make you not conclude the obvious thing?" produced four exceptions with clear conditions attached. Each one converts directly into a line on a card.

Session 4, scribe pattern. The apprentice drafts the procedure from the first three sessions. The expert reads it and makes nine corrections in about 25 minutes, four of which are steps the apprentice missed entirely and five of which are ordering fixes. Attempting the same document from a blank page had failed twice in the previous year.

Result. Four hours of paid time produced a two-page card with an ordered check sequence, four named exceptions, and five common errors. Twelve weeks later the second-best tech's first-visit resolution rate on that fault family had closed roughly half the gap to the expert. It did not close all of it, and it will not, because some of the gap is pattern recognition that only reps build. Half the gap on a fault family that recurs weekly, for four hours of investment, is a very good trade.

The arithmetic worth noting: the expert was about three times faster on the diagnosis step. Bringing one other tech from that gap to roughly half of it does not just save time on the calls the second tech takes, it removes the constraint where every hard one had to wait for one person's calendar.

What changes the answer

The expert is leaving in weeks. Drop the slow methods. Run the error catalog and the contrast interview back to back, in as many sittings as you can get, and record them with permission. Depth is a luxury you no longer have, and coverage of the rare high-consequence items is what matters. See related: Training the Tech Who Takes Over Your Tools.

The skill is physical rather than diagnostic. Feel and technique resist verbal capture much harder than reasoning does. Shift the weight to method 2 with the expert's hands on the apprentice's hands where that is safe, plus short video of the motion. Words will not carry it.

The expert is also your busiest producer. Do not schedule four separate hours off the calendar. Attach method 1 to jobs they were going to run anyway and accept a slightly longer job. The marginal cost is minutes, not hours.

A crew where several people are strong. Skip the interview format and run a case review with three or four of them on the same fault. Experts disagreeing with each other surfaces buried rules faster than any interview technique, because each one has to defend a step they had never had to justify.

How to verify you captured something real

• Hand the card to a mid-level tech with no explanation and watch them run the job. If they can follow it without asking a question that the card should have answered, it holds. If they ask, the answer to their question is the missing step. Add it and repeat.
• Check whether the card names failure modes and exceptions, not just steps. A card that is only an ordered list has captured procedure, which was probably already available. The exceptions are where the expert's actual value lives.
• Watch the expert read the finished card. If they say "that's not quite how I do it" and cannot say more, you are not done. If they can point at the specific line that is wrong, you are close and one more scribe pass will finish it.
• Measure something. First-visit resolution on that fault family, or the count of times the hard one waited for one person's calendar. If neither number moved in a quarter, the card is sitting in a folder and the capture was academic.

References

• Trade-standard practice for knowledge transfer and succession planning in skilled trades
• U.S. Department of Labor, on-the-job learning and mentorship structure in registered apprenticeship
• See related: How to Have a Trainee Teach It Back to You, Mentoring vs Managing: The Difference
• See related: Building an Apprentice's Diagnostic Vocabulary Over Time, The Difference Between Supervision and Mentorship