You get knowledge out of employees' heads by capturing it where it already shows up: in the questions people ask, the decisions they make and the work they hand over, and then writing it down as short, linked answers that the whole team (and your AI) can read. Asking experts to "document everything" rarely works. Turning repeated questions and real decisions into small pages does.

Why the knowledge is stuck

Some knowledge is easy to write down: a price list, a login procedure, a contract template. Researchers call this explicit knowledge. The harder part is tacit knowledge: experience, judgment and know-how that people use without thinking about it. The philosopher Michael Polanyi summed it up as the observation that we know more than we can tell (Polanyi, The Tacit Dimension, 1966, via Wikipedia). Skilled people often don't realize what they know, or that it would be valuable to someone else.

Dorothy Leonard and Walter Swap called the most valuable form of it "deep smarts": the judgment experienced people use to see the whole picture and spot the problem others miss (Harvard Business Review, 2004). That is exactly the knowledge a company loses when someone leaves, goes on holiday or simply becomes too busy to answer.

The cost shows up every day. In Atlassian's 2025 survey, 56% of respondents said the only way to get key project information is to ask someone or hold a meeting (Atlassian State of Teams, 2025). When the knowledge lives in one head, that head becomes a queue.

A useful model: the SECI spiral

Ikujiro Nonaka and Hirotaka Takeuchi described how knowledge moves between people and documents in four modes, known as the SECI model (Nonaka & Takeuchi, The Knowledge-Creating Company, 1995):

Mode	From → to	What it looks like at work
Socialization	tacit → tacit	Shadowing a colleague, pairing on a client call
Externalization	tacit → explicit	Writing down why you priced a project the way you did
Combination	explicit → explicit	Linking that note to the pricing page and the client
Internalization	explicit → tacit	A new person uses the page, does the work and builds their own judgment

Most companies are decent at socialization (people learn by sitting next to each other) and at combination (they collect documents). The step that usually fails is externalization: getting the "why" out of a person's head and into words. The practical methods below are mostly about that step.

Step 1: Start from questions, not from a blank wiki

Don't ask your best people to write a handbook. Ask them to answer the questions that already reach them.

For two weeks, note every question an expert gets in chat or in person.
Each question that comes up twice becomes a short page: the question as the title, the answer in a few sentences, an example.
Next time, the expert sends the link instead of retyping the answer.

This keeps the effort tiny and guarantees the pages are useful, because someone already needed them. How to Stop Answering the Same Questions Over and Over goes deeper on this habit.

Step 2: Capture decisions with their reasons

The most valuable and most often lost knowledge is why something was decided. A task tells you that the deadline moved; it rarely tells you that it moved because the client changed scope.

Make a simple rule: every decision that changes a goal, a price, a process or a customer relationship gets two or three sentences on the related task or page: what was decided, why, and who decided. Months later, that note is what saves a new person from re-arguing the same point.

Step 3: Use handovers and departures

Every handover is a chance to externalize knowledge, because someone else now needs it. Exit interviews and debriefs are a well-known way to capture knowledge from people who are leaving (Wikipedia, "Organizational memory"), but don't wait for a resignation:

Holidays: before leaving, the person writes a one-page "what you need to know" for each active project.
Role changes: the outgoing person records the five questions they get most and the answers.
Project ends: a 30-minute review: what worked, what didn't, what we'd do differently. Write it down and link it to the client.

Step 4: Interview, record and let someone else write

Some experts will never write. That's fine. A 20-minute conversation, recorded or transcribed, is often enough. Someone else (or an AI) turns the transcript into a draft page, and the expert only corrects it. Correcting is much faster than writing from scratch, and it gets the tacit "it depends" parts that people forget when they write.

Good interview questions:

"Walk me through the last time this went wrong."
"What do new people always get wrong here?"
"How do you decide between option A and B?"

Step 5: Link knowledge to the work

A page that sits in a folder gets forgotten. A page that is linked to the process, the client and the active project gets read, because people land on it while working. Keep pages small, one topic each, and link them to each other and to the tasks they apply to.

Step 6: Keep it current with light ownership

Knowledge rots. Give each important page an owner and a natural moment to check it, for example when the process runs again or a client contract renews. Don't build a review bureaucracy; just make it easy to see which pages haven't been touched in a long time.

What AI changes

AI makes two of the hard steps cheaper:

Writing: turning a call transcript or a long chat thread into a clean draft page takes seconds.
Finding: people no longer need to know where a page lives. They ask, and the AI answers from the company's own pages, with a link to the source.

It doesn't remove the need to capture knowledge in the first place. An AI can only answer from what has been written down or recorded somewhere it can read. See How to Give ChatGPT or Claude Access to Company Knowledge for the options.

How Orakis does it

In Orakis, knowledge lives next to the work. The Knowledge area holds processes, decisions, customers and people as linked pages, with a graph of how they connect, and goals, tasks and team chat sit in the same workspace. Alfred, the built-in AI agent, reads every page and answers questions from it, so an expert can point people to Alfred instead of answering again. Claude, ChatGPT and other assistants can read the same knowledge through MCP. Orakis is a young product in closed beta, so expect it to keep changing.

FAQ

How do I get experts to share without slowing them down? Ask them to answer real questions once and send a link next time. Offer to write the page from a short conversation so they only need to correct it.

Can all tacit knowledge be written down? No. Some judgment is only learned by doing, which is why shadowing and pairing still matter. Writing down the reasons behind decisions captures a large share of what usually gets lost.

Where should we start if nothing is documented? With the ten questions your team asks most often. Write those ten pages this month and link them where people work.

How to Get Knowledge Out of Employees' Heads