Knowledge management and knowledge base: what's the difference?

Knowledge management is a practice: deciding what you keep, who is answerable for it, how you keep it up to date. The knowledge base is the place where that knowledge is stored and searched. You can have a base without the practice: it fills up, then goes stale; and a practice without a base, which holds up as long as the team is small.

Where to start in an SME?

With a week of notes: the questions asked and received, exactly as they came. Then one owner per topic, and the answers written down at the moment they're given, with a date. Don't start by choosing a tool, and don't start by writing procedures cold: those are the two surest ways to finish nothing.

Do you need knowledge management software?

Not to begin with. You need one the day the volume goes beyond what one person can recall from memory, or when the knowledge is scattered across several places. A good tool searches what already exists and cites its source, rather than asking you to rewrite everything inside it; and it captures new answers at the moment they're given.

How do you capture the knowledge of an employee who's leaving?

Ask them for the list of questions they get asked, name who takes over each topic, have them write the answers down while they are fresh, date them, and set up a pairing over the notice period for what can't be written down. Accept that part of it will be lost: what you keep is what was written before, not during.

How do you keep a knowledge base up to date?

One owner per topic, a date on every answer, a review date, and the right to remove. With us, an answer only enters the knowledge base once it has been validated, and a wrong answer is corrected in one place, for everyone. Software documentation, for its part, re-syncs on its own when the vendor publishes.

Knowledge management: what walks out when someone leaves

Published on 30 October 2026

What is knowledge management?

Knowledge management is the set of practices that let a company retain what it knows, find it again when it is needed, and pass it on when people change. It is also called knowledge capitalisation when the emphasis is on what gets built up. The term is off-putting, because it brings to mind empty wikis and procedures nobody reads; the practice is simpler than that, and it starts by looking at what people ask a colleague ten times a day.

The moment you find out you've managed none of it is when someone leaves. The person who knew how to close a particular file, who to write to at a given vendor, why things are done this way, walks out with all of it in their head. What follows holds for a company of ten people as much as for one of two hundred, and needs no tool to get started.

The three layers of a company's knowledge

Everything a company knows fits into three layers, and almost every knowledge management project fails because it deals only with the first.

The level	An example	Where it lives today	What wastes it
What is written down	Procedures, instruction sheets, templates, minutes, software documentation.	In shared folders, a document management system, an intranet, vendors' websites.	Nobody can find it, or it's out of date and nobody knows.
What has been answered once	The answer from a software vendor's support, the email from a colleague who explains, the fix found one Friday evening.	In a mailbox, a chat thread, the head of whoever asked.	The same question comes up again a fortnight later, with someone else.
What is known without ever having been written down	The knack, the exceptions, a client's history, who knows what in the company.	In people's heads, and nowhere else.	Someone leaves, someone is off for months, a desk moves to another floor.

A five-step method, no tool required

First step: for one week, write down the questions you're asked and the ones you ask, exactly as they come. No sorting, no formatting. At the end of the week, the list tells you what deserves to be kept, and it's rarely what you'd have guessed.

Second step: for each topic on the list, name the person who has the final say. Not the one who knows the most, the one who decides when two answers contradict each other. Knowledge with no owner goes stale without anyone noticing.

Step three: write the answer at the moment you give it, not before. A procedure written in the abstract, for the sake of it, comes out long and wrong; an answer written in the moment, for a real person, comes out short and right. The question becomes the title, in the words of whoever asked it.

Fourth step: date every answer and set a review date. A 2024 answer about software that has changed twice since is worth less than no answer at all. The owner reviews it on the date, or removes it.

Fifth step: a single place to look. If part of the knowledge sits in the shared drive, part in email and part in a wiki, nobody looks anywhere and everyone asks the desk next door. Which place it is matters less than there being only one.

What an internal assistant adds to it

The five steps hold without any software. What a tool changes is the cost of each one. With us, the knowledge base isn't a wiki to fill in: the assistant searches what already exists, the procedures you've uploaded, the software documentation, the shared documents, and answers by citing the passage. The first level becomes findable without anyone rewriting it.

The second level enters the knowledge base the moment it is created. A question with no answer goes, if you decide so, to your subject expert, your IT person or vendor support; the answer that comes back is checked by an authorised person, in a single To process queue, and joins the knowledge base with its date. That's the third and fourth steps, done in passing, and this relay is live with our clients.

The third level, though, can't be written down, and we don't claim to capture it. What the assistant can do is say who: everyone lists their skills on their team profile, the answers they have validated show up there, and when a question needs a human, the assistant names the person who has already handled a similar case. The journal, finally, keeps the life of the company, what was decided and why, where an email would have lost it.

What leaves when someone leaves

The day someone announces they're leaving, a few weeks remain. The method above applies at speed: ask them for the list of questions people bring them, name who takes over each subject, have them write the answers down on the spot while they're still giving them, date all of it. What can't be written down is passed on by pairing someone with them over that period, and you accept losing part of it.

What changes with an assistant that has lived in the company before the departure: the answers that person validated are already in the knowledge base, with their name on them, and the questions people asked them have left a trace. You do not rebuild everything in three weeks, you re-read what is already there.

The questions we get asked

Knowledge management and knowledge base: what's the difference?: Knowledge management is a practice: deciding what you keep, who is answerable for it, how you keep it up to date. The knowledge base is the place where that knowledge is stored and searched. You can have a base without the practice: it fills up, then goes stale; and a practice without a base, which holds up as long as the team is small.
Where to start in an SME?: With a week of notes: the questions asked and received, exactly as they came. Then one owner per topic, and the answers written down at the moment they're given, with a date. Don't start by choosing a tool, and don't start by writing procedures cold: those are the two surest ways to finish nothing.
Do you need knowledge management software?: Not to begin with. You need one the day the volume goes beyond what one person can recall from memory, or when the knowledge is scattered across several places. A good tool searches what already exists and cites its source, rather than asking you to rewrite everything inside it; and it captures new answers at the moment they're given.
How do you capture the knowledge of an employee who's leaving?: Ask them for the list of questions they get asked, name who takes over each topic, have them write the answers down while they are fresh, date them, and set up a pairing over the notice period for what can't be written down. Accept that part of it will be lost: what you keep is what was written before, not during.
How do you keep a knowledge base up to date?: One owner per topic, a date on every answer, a review date, and the right to remove. With us, an answer only enters the knowledge base once it has been validated, and a wrong answer is corrected in one place, for everyone. Software documentation, for its part, re-syncs on its own when the vendor publishes.

To see it

The demo of a manufacturing company is open without signing up. Ask what the limit is for a meal while travelling: the assistant answers and cites the procedure. Ask who can help you with the till: it searches the team's skills. Try the demo, or first read why an internal AI.