How to turn your best employee's knowledge into written procedures before they leave
A repeatable interview method plus AI transcription and drafting turns one person's tribal knowledge into SOPs and checklists in a few working sessions.
- A structured interview, recorded and transcribed, is the core of the capture.
- AI drafts the procedures; a human who does the work checks every step.
- Plan for several one-hour sessions spread over a few weeks, not one marathon.
- Someone outside the work, like an AI Concierge, can run the capture with you.
The moment usually arrives in a sentence. The one employee who knows how the scheduling really works, which vendor account has the quirky payment terms, and how to calm the customer nobody else can handle says they are leaving or retiring in a few months. The owner nods, and then realizes almost none of it is written down anywhere. It lives in one person's head, and that person has a last day on the calendar.
An interview-plus-AI capture process turns that knowledge into written procedures before the last day arrives. You record structured conversations with the departing employee, let AI transcribe them and draft the procedures, then have someone who does the work verify each step. This article lays out the full method, the prompts to use, and where outside help fits.
What the capture process does
The process has three moving parts. First, a structured interview, which means a planned set of questions rather than a casual chat, walked through task by task. Second, AI transcription and drafting, where a tool like ChatGPT or Claude turns the recorded conversation into a clean first draft of a procedure. Third, verification, where the departing employee or a colleague confirms the draft matches what actually happens. The output is a set of standard operating procedures, called SOPs, which are simply step-by-step written instructions, plus checklists others can follow without asking questions.
The reason to interview instead of asking the employee to write everything down is simple. Most people cannot write their own work down well, and most will not get to it. But almost everyone can talk through what they do for an hour, especially when someone else asks the right questions. The interview gets the knowledge out; AI does the writing.
Who this is for, and who it is not for
This fits an owner-led business where one person holds operational knowledge that would hurt to lose. That is common in home services companies, clinics, professional services firms, real estate offices, and restaurants across Texas, where a long-tenured office manager or lead technician often holds the place together. If the person has already given notice, or has announced a retirement date, the process is worth starting this week.
It is not for capturing knowledge the employee does not actually have, or does not want to share. If someone resists, the interviews stall, and no tool fixes that. It also does not replace training. A written procedure gets the knowledge out of one head and onto paper; the replacement still has to practice the work.
How to run the capture
- 01List the knowledge at risk
Write down every task, account, and relationship only this person handles.
- 02Rank by pain
Put first the items that would break the business fastest if nobody knew them.
- 03Book one-hour interviews
Schedule one hour per topic block, a few days apart, before the last day.
- 04Record each session
Record the interview with the employee's consent, and let them talk through the task while doing it if possible.
- 05Transcribe and draft with AI
Feed the transcript to ChatGPT or Claude with a prompt like: turn this transcript into a step-by-step SOP, list every tool and login mentioned, and flag anything unclear.
- 06Verify every draft
Have the departing employee, or a colleague who does similar work, correct the draft until each step matches reality.
- 07Store and assign
Save the finished SOPs where the team can reach them, and assign each one an owner who will keep it current.
Two interview habits make the drafts much better. Ask the employee to narrate the task in order, including the parts that feel obvious, because those are the steps everyone else is missing. And ask about exceptions, what they do differently when the normal path fails, since exceptions are where most written procedures fall apart.
A useful prompt for the drafting step is to paste the transcript and write: this is a transcribed interview about how we handle [task]. Write an SOP with numbered steps, a list of every system, account, and contact mentioned, and a short section titled edge cases for anything unusual the speaker described. Ask me about any gaps before finishing. That last line matters. It pushes the AI to surface what the interview missed, which becomes your question list for the next session.
Tools and costs
You need a way to record, a transcription option, and an AI chat tool. Many owners already carry a phone that records, and the major AI assistants can work from pasted transcripts. Owners usually need one AI subscription such as ChatGPT or Claude, paid directly, and any cost is worth naming before you buy. Beyond the subscription, the method costs management time: a few hours of interviews per topic, plus review time for each draft.
The whole capture typically runs across a few weeks. Two to four one-hour sessions is a realistic start for a single role's core tasks, and more if the person wears many hats. Spreading the sessions out matters, because the gaps you find in session one become the questions for session two.
Where outside help fits
An owner can run this alone, and many will. The structure above is the whole method. Where it usually breaks down is time and follow-through. The interviews get postponed, the drafts sit unverified, and the last day arrives anyway. That is the situation where outside help earns its place.
At Texas AI Lab, an AI Concierge can run this capture process with you. The Concierge is a real person from the team, not a chatbot, who learns your business and works alongside you, with two 1-hour working calls a month plus voice messages and email between calls. The first month builds a personal context file in your own AI account, and everything built lives in your accounts and is yours to keep. A 25-minute fit call on Zoom, booked through the contact page, is the place to find out whether that fits.
