06 — FACILITATION METHODOLOGY: Knowledge Transfer
How to run the capture session — so the source recording actually contains the judgment, not just the steps. The engine can only mine what the recording holds; this is where the quality is won or lost.
This is the AI-native form of the KTP interview (Exercises 1–3) plus Educate & Delegate ("record the task as it naturally happens").
The Principle
The steps are the easy part and usually already written down. The value is the judgment — the calls the expert makes without thinking, the "I'd never do X before Y," the "this client always…". Those don't surface if you ask "what are the steps?" They surface when the expert is doing the work and narrating the calls. So: capture them doing it, and prompt for the judgment.
Tools: screen-share + recording (Zoom/Loom) + transcription (Otter). Keep each recording focused — one task/process per recording, ideally under ~20 minutes (Educate & Delegate's rule). Several short, real recordings beat one polished marathon.
Before the Session — Prioritize (KTP Exercise 1)
Pick the subject by risk, not convenience.
- Who is transitioning, or whose knowledge would hurt most if it walked out the door tomorrow?
- Which process, if it broke, would a client notice first — and how fast?
- What does this person handle that no one else knows how to do?
Capture the one with the highest risk × nearest deadline first. It's rinse-and-repeat — you'll run it again.
During the Session — Record the Work + Pull the Judgment (Exercise 2)
Have the expert do the actual task on a shared screen and narrate it. Don't let them abstract into "well, generally…" — keep them in the real instance. As they go, prompt for the four things the engine needs for each section:
| Section the engine fills | Prompt that pulls it |
|---|---|
| Set (standing setup) | "Before you start — what has to be true/current for this to work? Which accounts, tools, logins?" |
| Rules | "Anything that always goes the same place? A vendor, a category, a routing rule?" |
| Standing notes / watch-outs ← the heart | "What just made you pause? … Why that and not the obvious choice?" · "What would a new person get wrong here?" · "What do you check that isn't in any instructions?" |
| Exchanges | "What do you have to send or chase each cycle, and from whom?" |
| Outputs | "What does 'done' look like? What won't you ship until it's true?" |
The watch-out prompts are the job. Every time the expert hesitates, corrects course, or says "well, except when…", stop and capture why. That hesitation is tacit knowledge becoming explicit.
Watch for the "I don't want to be replaceable" reflex (it's normal — name it if it shows up): the expert under-shares the judgment that makes them valuable. Reassure that the goal is continuity, not replacement, and that the best transfers come from the people who did the work best.
Close the Session — Recommendations (Exercise 3)
Ask the departing expert directly:
- "If you were training your successor, what would you make sure they understood first?"
- "What's the one thing that, done wrong, causes the most damage?"
- "What's not written down anywhere that should be?" → these become
FLAG:: RISK —items.
After the Session — Hand to the Engine
- Clean the transcript (fix obvious mis-transcriptions; don't rewrite the substance).
- Run the kit (
05): mine the transcript into the content file, flag gaps, generate, gate. - Anything the recording didn't cover stays a
FILL::flag — that's the honest output, and it tells you exactly what to capture in the next session.
A gap-heavy first capture is a signal, not a failure — it shows precisely where the knowledge is still trapped. Run another focused recording on the flagged areas and Update.