Knowledge capture interview guide
For the first process test, run by hand. Go through the passes in order with one expert and one real process. Ask one question at a time and write each answer straight into the step cards.
Before you start
Agree these four things with the company first. A session should last about 30 minutes; split passes 3 and 4 into a second session if the process is long.
One process
that repeats, depends on one person, and matters to the business. Not a one-off.
One expert
the person who does it best today.
One beginner
who has never done it, for the test afterwards. Don't let them watch the capture.
A recent real case
ask the expert to think of the last time they did it, and to bring any form, template, file or screenshot they use.
Ground rules
- 1Ask one question at a time, then stay quiet. Silence makes experts add the detail they skipped.
- 2Ask about a real case (“the last time”), not the general rule. People remember what they did far better than what they do in general.
- 3Don't finish their sentences and don't suggest answers. Write their words, not yours; keep exact quotes for anything unclear.
- 4Experts skip steps that feel obvious. When a step sounds too short, ask what happens in between.
- 5Let them speak Darija, French or Arabic. Write notes in their language, with a short note in yours if needed.
- 6Tag every line in your notes: said (their words), inferred (your guess, to confirm) or seen (you watched it). Never publish an inferred line unconfirmed.
Opening line
“I'd like to understand how you do this job, step by step, as if I had to do it tomorrow. I'm not testing you. The goal is a guide a new person could follow without asking you. Everything we write, you will check and correct.”
You can say
We write down how you work, you validate every part, and we test whether another person can follow it.
Don't say
That it removes training, that the AI understands their company, or any time or cost figure.
PASS 0· about 5 min
Scope
Name the process, what starts it, who does it, and what already exists.
| # | Ask | Listen for |
|---|---|---|
| Q1 | “Which process are we capturing? Give it a name a new person would recognise.” | One process with one outcome. If it sounds like three jobs, pick one. |
| Q2 | “What starts it? What tells you it's time to do it?” | The trigger: a request, an event, a date, a message. |
| Q3 | “What is the result when it's done well, and who receives it?” | The outcome and the person or team who gets it. |
| Q4 | “Who does this today, and who should be able to do it after we've finished?” | The expert and the beginner's role. |
| Q5 | “Is anything written already: a procedure, form, checklist, template, screenshot? Can I see it?” | Existing material. Ask what in it is out of date. Collect copies. |
| Q6 | “How often does it happen, and how long does one run take?” | Frequency and duration, as numbers. |
PASS 1· 10 to 15 min
Real-case walkthrough
Get the true sequence of one real run, with its context. Don't interrupt the first telling. Number each step as they say it, in their words.
| # | Ask | Listen for |
|---|---|---|
| Q7 | “Think of the last time you did this. Walk me through it, from what started it until it was finished.” | Order of events, names of tools, files and people. Write it as a numbered list of steps. |
| Q8 | “Was that run typical, or was something unusual?” | What was different. Unusual runs reveal the exceptions. |
| Q9 | “How did you know it was finished?” | The real finish line. Often not the one in the official procedure. |
Neutral follow-ups
Use these only where they skipped something, one at a time.
- “What happened next?”
- “Where were you, and what were you looking at: which screen, form, paper or message?”
- “Who else was involved? Did you wait for anyone?”
- “What did you decide at that point?”
- “What did you do between those two things?”
When the walkthrough ends, read the numbered steps back and ask: “Did I miss anything between these?” Then continue to pass 2.
PASS 2· about 10 min
Coverage
Fill the process card (see the sheets at the end). Check which fields are still empty, then ask about the biggest gap first, one question at a time. Skip any field already answered in pass 1.
| # | Fills | Ask | Listen for |
|---|---|---|---|
| Q10 | Purpose | “Why does the company do this? What goes wrong if it isn't done?” | The reason in business terms, and the cost of failure. |
| Q11 | Inputs | “What do you need in hand before you can start? Where does it come from?” | Documents, data, approvals, materials, and who provides them. |
| Q12 | Outputs | “What exactly do you produce or hand over, in what form, and to whom?” | A concrete deliverable, not “it's handled”. |
| Q13 | Success criteria | “How do you and others tell it was done well, not just done?” | Something checkable: a number, a status, a customer reaction. Vague words (“good”, “fast”) need a follow-up. |
| Q14 | Tools | “Which systems, files, forms, phones or people do you use along the way?” | Names of tools, and what each is used for. |
| Q15 | Roles | “Who else touches this? What does each person do, and what do they wait for?” | Handoffs and waiting points. |
| Q16 | Timing | “Is there a deadline or a time by which it must be done?” | Time limits and what happens if they are missed. |
PASS 3· 3 to 5 min per step
Step deepening
A step is good enough when a stranger can answer seven things: what do I do, why, what do I need, what do I produce, how do I know it worked, what can go wrong, and what do I do then. Go through the steps from pass 1 in order and ask these for each one.
| # | Ask | Fills on the step card |
|---|---|---|
| S1 | “What exactly do you do here? What do you touch, click, write or say?” | What I do |
| S2 | “What are you looking at or checking at this point?” | What I do (the check) |
| S3 | “What information do you need here, and where do you find it?” | What I need |
| S4 | “Why do you do it this way? What goes wrong if it's skipped?” | Why (key points and reasons) |
| S5 | “What do you have at the end of this step?” | What I produce |
| S6 | “How do you know this step worked?” | How I know it worked |
| S7 | “What do beginners get wrong here?” | What can go wrong |
| S8 | “What do you notice here that a beginner wouldn't?” | Key point (the hidden skill) |
| S9 | “If the expected thing isn't there, what do you do?” | What to do if it goes wrong |
| S10 | “When would you stop and ask someone? Who?” | Escalation |
| S11 | “Can I see it, or can you photograph the screen or form?” | Evidence (photo, file) |
Vague
“Check the invoice.”
Precise
“Compare supplier, quantity and price on the invoice with the purchase order. If anything differs, don't approve it and notify procurement.”
If your note for a step still reads like the first one, you are not finished with that step.
PASS 4
Judgment probes
Turn judgment into rules a beginner can follow. Listen in every pass for these phrases, and stop to probe each time you hear one.
- “it depends”
- “I check”
- “use judgment”
- “usually”
- “normally”
- “just”
- “of course”
- “you know”
- “enough”
- “properly”
- “quickly”
| # | Ask | Aim |
|---|---|---|
| P1 | “What does it depend on?” | List every factor they name. |
| P2 | “What tells you which way to go?” | The signal: something seen, a number, a customer statement. |
| P3 | “Think of the last two times. What was different when you chose differently?” | Real cases expose the criteria they can't state in general. |
| P4 | “If a beginner had to decide, what rule could we give them? Is there a number or a limit?” | A checkable threshold. |
| P5 | “What would a beginner miss in that situation?” | The hidden cue. |
| P6 | “When do you break that rule? Who can approve an exception?” | Exceptions and who decides. |
| P7 | “When you say “enough” or “properly”, how do you know?” | Replace vague words with a test. |
Write each answer as a rule: if [condition] then [action], otherwise [action]. Read the rule back and ask: “Is that how you'd decide?” If they hesitate, there is still a missing condition.
PASS 5· about 10 min
Teach-back
The expert checks your draft claim by claim. Rewrite your notes as a guide a beginner would read: numbered steps, with the why, the checks and the rules. Then read it to the expert one step at a time.
| # | Ask | Record |
|---|---|---|
| T1 | “Is this what you do? Anything wrong?” | For each claim: confirmed, edited or rejected. |
| T2 | “Is anything missing between these two steps?” | New steps or conditions. |
| T3 | “If a new person followed only this page, where would they get stuck?” | Predicted gaps. |
| T4 | “What would you still have to tell them out loud?” | The strongest signal of what is still missing. Write it down word for word. |
Count the corrections the expert makes (edited plus rejected claims). It tells you how good the first draft was, and it goes into the scorecard. Leave the guide with the expert to check once more before the beginner sees it. Only a version the expert has confirmed goes to the beginner.
After the session
The capture only counts if a beginner can use it. These three steps prove it.
1. Dry run (optional, about 15 min)
Paste only the confirmed guide into an AI chat with this prompt, and turn each point it lists into a question for the expert. This is weaker evidence than a real person and never replaces the beginner test.
“You are a new employee who has never done this work. Using only this guide, list every point where you would get stuck or would have to ask someone, and say why.”
2. Beginner test
Give a beginner who has never done the process the confirmed guide and a real task.
- 1Give them only the written guide. The expert is not in the room, but can be reached.
- 2You stay silent and observe. Don't hint, even when they struggle.
- 3Log every hesitation, every question and every time they ask the expert (see the beginner test log).
- 4Note the time to finish, the result, and any error or deviation from the guide.
- 5Ask the beginner afterwards: “Where was it unclear? What did you have to guess?”
3. Gap follow-up (about 20 min)
Take the logged gaps back to the expert as the only agenda. Update the guide, save it as version 2, and test it again with a beginner (the same person only if they agree, since they now know the process).
Reading the result. Compare the expert interventions with your “before” from the discovery interview (how often seniors are interrupted for this process). Success is clearly fewer interventions, with the process completed. If the beginner still needs the expert often, the log shows exactly which knowledge is missing. That is a useful result, not a failure.
Capture sheets
Copy these tables for each process. Tag every answer said, inferred, seen or confirmed (after the expert agrees in pass 5).
Process card
| Field | Answer | Tag |
|---|---|---|
| Process name | ||
| Purpose (why it exists, what fails without it) | ||
| Trigger (what starts it) | ||
| Outcome and who receives it | ||
| Inputs needed before starting | ||
| Roles and handoffs | ||
| Tools, systems, files | ||
| Success criteria (how we know it was done well) | ||
| Frequency and duration | ||
| Expert, and beginner's role | ||
| Existing documents (and what is outdated) | ||
| Language the expert used |
Step card
One row per step; add rows as needed.
| # | What I do | Why (key points) | What I need | What I produce | How I know it worked | What can go wrong, and what to do | Escalate to / when |
|---|---|---|---|---|---|---|---|
| 1 | |||||||
| 2 | |||||||
| 3 | |||||||
| 4 | |||||||
| 5 |
Decision rules
From pass 4.
| Step | If (condition) | Then (action) | Otherwise | Who can approve an exception |
|---|---|---|---|---|
Beginner test log
One row per hesitation, question or help request.
| Step | Where they hesitated or got stuck | Question they asked | Did the expert help? (Y/N) | Error or deviation from the guide | Minutes |
|---|---|---|---|---|---|
Working definition of Expert Assistance Rate: steps where the beginner needed the expert, divided by total steps. Also record the plain count of expert interventions. Use the same two numbers for version 1 and version 2.
Scorecard
| Metric | Version 1 | Version 2 |
|---|---|---|
| Expert corrections in teach-back | ||
| Expert interventions (count) | ||
| Steps that needed the expert, out of total | ||
| Completed? (yes, partly, no) | ||
| Time to finish (minutes) | ||
| Errors or deviations | ||
| Beginner confidence (1 to 5) | ||
| Expert confidence in the guide (1 to 5) |