Ask an entrepreneur to write down his way of working and you get a four-page document. Then ask him how he put together that one quote from last week, and you get twenty minutes of reasoning that isn't in any of those four pages.

That difference is the subject of this piece. Knowledge management, capturing and keeping usable the knowledge your business runs on, doesn't fail out of unwillingness. It fails because the form we try it in is the wrong form.

01 / the patternWhy does knowledge leak away?

Because it never settles anywhere. Every working day you make dozens of small decisions that land nowhere. Why this client gets an exception and that other one doesn't. Which question you ask when an intake goes too smoothly. When you'd be better off turning down a job.

Those decisions are your trade. They're also exactly the part your systems don't see. A CRM records that a quote went out, not why it was put together the way it was. A time-tracking system knows a job took twelve hours, not that eight of those went into one exception you could have seen coming.

Your systems record what's finished. Your knowledge sits in what led up to it.

The result is a business that runs on paper and in reality hangs on you. Everyone can carry out the steps. No one can make the trade-offs. So everything that deviates comes back to you, and in knowledge work deviating isn't the exception but the bulk of the work. That pattern itself, separate from the question of how you capture knowledge, is described in you don't have a business that runs.

02 / the typesWhich knowledge should you capture?

Not everything. That's the first bend where these kinds of attempts fly off the road: people start at the beginning and write a manual. That takes months and doesn't get read, because it's full of things everyone already knows.

Knowledge in a business falls into three types, and each needs different treatment.

Anyone who wants to capture knowledge and starts at row 1 is doing the work with the least payoff. The pain sits in rows 2 and 3, and those two don't roll out of a template. They come out when someone probes further.

03 / why it goes wrongWhy doesn't a manual work?

For three reasons, and they stack.

You don't know what you know. Expertise has the property of becoming invisible as it grows. What you've done for fifteen years doesn't feel like knowledge but like something obvious. Exactly those obvious things are what a new employee misses, and you don't mention them by yourself, because you no longer see them.

Writing is a different trade than doing. Explaining is easier than writing down. In a conversation you make associations, give examples, correct yourself halfway. Behind a blank document you don't do that. You write the tidy version, and the tidy version misses the exceptions.

It's never finished. A manual is done on the day your way of working changes, and that changes continuously. So maintenance grinds to a halt, and after a year no one trusts the document any more. A document that isn't trusted doesn't get used, and then the time put into it was wasted.

This isn't a discipline problem. It's a form problem. The solution isn't to try harder to write it down, but to choose a different form in which the knowledge does come out.

04 / the methodSo how do you get it out?

By talking and recording it. Not by filling in a questionnaire.

The difference is bigger than it sounds. A form asks what you think is important. A conversation in which someone probes further exposes what you no longer see as special yourself. The question that does the work isn't "how do you do this", but "why did you do it differently for that client".

The chain after that isn't complicated, and it's worth knowing how it runs before you buy it from someone:

1. Recording. A conversation of one to one and a half hours about one clearly defined part of your work. Not the whole business, or it becomes superficial.
2. Transcript. Written out word for word. This is deliberately the raw form: the nuance sits in the detours and self-corrections, which a summary strips out.
3. Knowledge layer. The transcript is pulled apart into separate, verifiable units. One rule, one source. That way you can later see, piece by piece, where something came from and what's no longer accurate.
4. Application. This is where software finally comes in. The knowledge layer feeds whatever you want to do with it: drafting documents, answering questions, preparing work.

For whom this pattern is most painful: that's the knowledge entrepreneur, and that group is bigger than the term suggests: a financial planner who has never written down his trade-offs falls under it just as much.

Step 3 is where it stands or falls. Knowledge kept as one big document can't be maintained or checked. Knowledge held in separate units with a source attached can be adjusted piece by piece when your way of working changes. That's the difference between an archive and a living system.

One rule, one source. What you can't trace, you can't trust.

And note the order. Software is at position four, not position one. Anyone who starts by picking a tool is choosing a form before knowing what needs to go into it. That's the order reversed. How to approach it properly is in where do you start with AI.

05 / the testHow do you know it's worked?

By one thing: less comes back to you.

That's the only measure that counts, and it's sharper than it looks. Not "we now have documentation", not "the system is up", but: the questions that landed on your desk last month don't land there this month. If that doesn't happen, knowledge has been captured that no one needed.

Two useful interim measures:

• Can someone else find the answer without you? Not "is it written down somewhere", but: does he find it, and does he dare to rely on it. That second part is a matter of trust, and you only win that with traceability.
• What happens with an exception? A system that handles the standard cases and stalls at every deviation has captured row 1 and left rows 2 and 3 behind. Then you've built a manual with better technology.

What you should not measure is how much is in it. Volume isn't quality. A knowledge layer of two hundred sharp units is worth more than one of two thousand vague ones.

06 / the limitWhen shouldn't you start on this?

There are three situations in which we'd say: don't do this now.

Your way of working isn't settled. If your offering changes substantially every quarter, you're capturing a moving target. Waiting is then cheaper than building. This applies to young businesses more often than they think.

You're the only one and that's staying that way. If you work alone and want to keep working alone, the return is small. The gain from captured knowledge sits in transfer, and without anyone to transfer it to, you're mainly paying for a tidier archive.

The problem is capacity, not knowledge. If you're stuck because there's simply too much work and too few hands, this is the wrong tool. Capturing knowledge helps against repeat work and dependency, not against an overfull schedule of work only you can do.

In all three cases we'd simply say so in a conversation. That's not modesty: building a system on a moving foundation produces a system that no one trusts within a year, and that's no use to anyone. Which bottleneck you do need to have before capturing pays off is in how do you know if you're ready for an AI system.

Further reading: why captured knowledge must remain yours is in whose is it when it works, and what such a system explicitly does not do is in what AI doesn't do.

This text was produced with AI assistance and checked and approved for publication by a human.