How to capture a process while the work is being done
Knowledge is most accurate at the moment someone uses it; the longer you wait, the more of it is reconstructed from memory. To capture a process while it happens, agree what to record, let the expert do a real case while explaining their choices, ask about the decisions straight after, turn the recording into a draft instruction, and have the expert confirm it before anyone uses it. A second, comparable case shows what the first one missed.
Before you start
Prerequisites: a willing expert who controls the recording — start, pause, stop — agreement on what may be recorded (screens, systems, customer data), and a representative case coming up soon. Screen and voice recording tools work well; so does a colleague taking careful notes.
Use this guide when the expert struggles to explain the work in an interview, when the work depends on screens, materials or machines, or when the steps look obvious to the person doing them and nobody else can repeat them. It works well alongside an interview: the interview gives the reasoning, the real case shows the practice.
The steps
Step 1 — Agree what to capture, and what not
Decide which task, which screens and files, and how long raw recordings are kept. Mask customer data where you can.
Pitfall: Recording everything “just in case”. It creates privacy risk and hours of footage nobody reviews.
Step 2 — Capture a real case, narrated
The expert does the work as usual and says what they look at and why. The observer only asks when a choice is made silently.
Pitfall: A staged demonstration. Experts skip the steps that feel obvious to them — exactly the ones a successor needs.
Step 3 — Ask about the decisions straight after
On the same day, go through the choices: what told you to do that? What would have made you do it differently? When does this not apply?
Pitfall: Waiting a week. The reasoning is already being tidied up in memory.
Step 4 — Turn it into a draft instruction
Structure the recording into steps, decisions, checks and exceptions. State-of-the-art AI tools can transcribe and draft; mark what was observed, what was said and what is inferred.
Pitfall: Treating the AI draft as the source. A draft can invent a step or miss a silent one.
Step 5 — Let the expert confirm it
The expert corrects the draft and signs off the version. They can strike anything captured from them.
Pitfall: Publishing before review.
Step 6 — Test with a second case
A colleague handles a comparable case with the instruction. What goes wrong is what the first capture missed.
Pitfall: Stopping after one case. One case is one path; the next case finds the exception.
An example
Take a metal processing company that wants to capture how its most experienced operator sets up a press brake for a new part. The operator records the setup of a real order and explains what they check on the drawing and on the material.
Straight after, the observer asks two questions: why was the first test bend made on an offcut, and why was the angle corrected before measuring? The answers — about how this batch of material springs back — are in no machine manual.
The draft instruction marks those answers as said, not observed. The operator corrects two steps and approves the version. A colleague then sets up the next similar part using only the instruction. The one thing that goes wrong becomes a new step, and the operator confirms it.
Where MentX comes in
MentX captures how your company really works: the knowledge of your key people and the processes they run. We offer both a hands-on service approach — mapping your processes and capturing the knowledge behind them, from meetings, workshops and a guided intake — and our own personality-aware knowledge base, with full source attribution, in your own Knowledge Space. We build on the tools you already use and add best-practice solutions where they fit.
Capturing in the moment keeps knowledge close to real work — through Connect and Deepen for an expert, and when you capture your operations for a process. Every statement keeps its source and its date.
Checklist
- Scope and data rules agreed before recording
- A real case, narrated by the expert
- Decisions discussed the same day
- Draft marks observed, said and inferred
- Expert confirmed the version
- A second case run to find what was missed
Questions people ask
Does the expert have to agree to being recorded?
Yes. They control start, pause and stop, and approve what is kept. If recording is not acceptable, a colleague taking careful notes works too.
Can AI capture the process on its own?
AI tools can transcribe, structure and draft. They cannot tell which silent choice mattered. That still takes a person asking why, and an expert confirming the result.
How many cases do we need to capture?
Enough to meet the common variants of the work. Start with one real case, test the instruction on a second, and add a case whenever a colleague meets a situation the instruction does not cover.
