When the expert leaves, the knowledge usually leaves too
Knowledge walks out the door, usually without a ceremony
A veteran retires. A project lead takes another job. A department gets reshuffled. The shortcuts they knew and the judgment they used when the manual was wrong leave with them. Most teams notice after the fact, when onboarding takes longer and the same mistakes show up again.
The part that is hard to keep is not the SOP. It is the tacit stuff: when to skip a step, who to call, what "looks off" in a file. Knowledge bases stall because everyone has other work, and updating a wiki is never the fire of the week.
Talk first. Write second.
If someone is leaving or changing roles, record the conversations you were going to have anyway. Walkthroughs. Office hours. The messy Q&A. Transcribe them. Then use a model to pull out steps, decision rules, and the exceptions they keep mentioning.
They talk the way they work. You do the distillation after. What you get might be a FAQ, a process note, or a training brief. Pick the shape your team will actually open.
Finding it later is the other half
Even when the notes exist, they sit in Drive, email, and a folder named Final_v3. Internal search is the unglamorous win: hook a model up to the documents and recordings you already have, and let people ask in plain language.
The limit is usually the pile, not the model. Scattered files give scattered answers. Clean enough files give useful ones.
When this is worth doing
Skip it if you cannot name a result. "Cut new-hire ramp by a month" is a result. "Improve knowledge sharing" is a poster.
It also helps if the work repeats, the knowledge exists in some form (talk, docs, tickets), and a human reviews the first drafts. AI is a first pass. The expert still has to say yes.
A small loop you can run this month
- Pick one person whose knowledge you would miss next quarter.
- Record a few structured conversations on the processes they own.
- Transcribe. Feed the transcript to a model with a tight ask: steps, decisions, rules of thumb.
- Sit with them and fix what it got wrong.
- Put the final note where people already search.
You do not need a platform for that. Tools you already pay for will do. Prove it on one expert before you buy anything with "knowledge" in the name.
The point is not to replace the person. It is so the next person is not starting from a blank Slack history.