Enterprise knowledge management is an upkeep problem, not a software problem

The short version

• Enterprise knowledge management covers four jobs: capturing knowledge, keeping it current, controlling access and delivering it where people work.
• Knowledge with no named owner and no review date goes stale, and staff stop trusting it after a few wrong answers.
• An AI assistant improves how knowledge is delivered, but it repeats stale content with the same confidence as current content.
• In an enterprise knowledge system, permissions should be enforced before retrieval, because filtering results afterwards still exposes content to the model.
• A small firm where people can still ask the person who knows usually needs a tidy wiki, not a knowledge management programme.

What does enterprise knowledge management actually cover?

Enterprise knowledge management covers four jobs: capturing what the organisation knows, keeping it current, controlling who can see it, and delivering it when someone needs it. A tool can help with each job. No tool does all four by itself.

Knowledge comes in two forms. Some is written down: policies, contracts, runbooks, product documents, tickets. Some sits in people's heads: why a supplier was dropped, which exception a client always gets, what really happens when the official process fails. Written knowledge is a maintenance problem. Unwritten knowledge is a capture problem, and it usually leaves with the person who holds it.

The word 'enterprise' changes the problem in one way. Beyond a certain size, nobody can say who knows what, so the organisation needs a system to do that job. Content also spreads across more tools, more teams and more access rules.

What are the four jobs of enterprise knowledge management?

• •Capture: move knowledge out of inboxes, chats and heads into a shared place.
• •Upkeep: give each item an owner and a review date, and retire what is out of date.
• •Access: make sure people see what they are entitled to see, and nothing else.
• •Delivery: get answers to people where they already work, whether that is search, chat or a support tool.

Why do enterprise knowledge management programmes fail?

They fail because nobody owns the upkeep. The launch gets a budget, a migration and plenty of attention. Maintenance gets none of those, so content starts ageing from the first day.

The failure is quiet. Search still returns results and the pages still load. A process changes, the page does not, and someone follows the old version. After a few of those experiences, staff stop trusting the system and go back to asking a colleague.

A knowledge base people do not trust can be worse than having none. Experienced staff route around it. New starters cannot tell which pages are wrong, so they are the ones it misleads.

What does a maintained knowledge base need?

A maintained knowledge base needs a named owner for each area, a review date on each page, one agreed home for each topic and a way to retire old content. None of this requires new software. It requires someone with time set aside to do it.

• •A named owner for each area. A person, not a team, so there is someone to ask when a page is wrong.
• •A review date on every page. When the date passes, the page is visibly flagged as unreviewed.
• •One home for each topic. Two versions of a policy will eventually disagree, and nobody will know which is right.
• •A way to retire content. Old pages should be archived, not left in search results.
• •A route for reporting errors. Anyone who spots a mistake should be able to flag it in one step.
• •A log of failed searches. Questions that return nothing show you what to write next.

How do you capture knowledge that nobody has written down?

Capture it as close to the work as you can, by asking the person who holds it and writing the answer down for them. Asking busy experts to write documentation rarely works, because it is never their most urgent job.

Short recorded walkthroughs help, as do structured conversations before someone changes role. Resolved support tickets and project retrospectives are also useful, because the reasoning is already on record. Transcripts and AI summaries can reduce the writing, but the expert still needs to check the result. A tidy summary of a misunderstanding is still a misunderstanding.

Is a wiki the same as enterprise knowledge management?

No. A wiki is a place to keep knowledge. Enterprise knowledge management is the practice of keeping that knowledge correct, permissioned and easy to reach. You can run a sound practice on a wiki, and a poor one on an expensive platform.

Cost of entry is rarely the barrier. Notion's pricing page lists Free, Plus, Business and Enterprise plans, and Confluence's lists free and paid plans, on terms that can change. The harder decision is who will keep the content right after launch.

Larger platforms typically add search across many systems, connectors and usage analytics. Those help when content already lives in many places. They will not review a single page for you.

How does AI change enterprise knowledge management?

AI changes delivery, not upkeep. An assistant built on retrieval can take a plain question, find relevant passages across several systems and write an answer with links to the source. Staff no longer need to know where something is kept or what it is called.

AI also raises the stakes on content quality. A search page shows a stale document with its title and date, so a careful reader can judge it. An assistant can turn that document into a confident sentence. If two versions of a policy are in the index, the model may answer from the old one, and the answer will read just as well.

So the useful order is content first, assistant second. An assistant should show its sources, say when it cannot find an answer and log what it failed to answer. That log is valuable, because it tells content owners exactly what is missing.

How should permissions work in an enterprise knowledge system?

Permissions should be enforced before retrieval, so a search only ever touches what that user may read. Filtering results afterwards is not enough, because the content has already reached the model.

Copying content into a search index creates a second access control surface, and it does not inherit the rules of the first. Permissions also change after indexing. People move teams, projects close and folders get reshared. The index has to be reconciled with its source systems on a schedule, or its permissions will drift out of date.

What does the evidence on knowledge management actually show?

The published evidence is mostly survey data, often from companies with a commercial interest in this market, so treat it as directional. It tells you what respondents report. It does not tell you what return your organisation would get.

APQC argues that knowledge management makes knowledge workers more productive and less stressed, drawing on a 2021 self-reported survey of 982 full-time knowledge workers. That is a useful signal, not a measured result for any single firm.

Larger samples do not remove a commercial interest, and a survey about AI is not evidence about knowledge management. We have no client results to offer either, and we will not borrow a figure to fill the gap.

The better measure is your own. Count how often the same question gets asked, how long a new starter takes to find an answer, and how many pages are past their review date.

Who was actually surveyed?

When is enterprise knowledge management not worth doing?

It is not worth doing when people can still get a correct answer by asking the person who knows, and that person is not overloaded. A small firm in that position needs a tidy shared drive and a short wiki, not a programme.

It is also not worth starting if nobody will own the upkeep. A launch with no owner produces a stale system and a reason to distrust the next attempt.

The case gets stronger when the same questions keep coming back, when answers are written down but spread across systems, and when one or two people have become a bottleneck. It also gets stronger when staff need to show which version of a policy they relied on.

How do you start enterprise knowledge management without a big programme?

Start with one area that generates repeat questions, fix the content there, and prove people use it before widening the scope. HR policies, support runbooks and onboarding material are common starting points, because the questions repeat and the answers usually exist somewhere.

1. Collect the repeated questions from chat, email and the support queue.
2. Pick one area and name its owner.
3. Find the best existing answer to each question, and archive the duplicates.
4. Write down what is missing, especially what only one person knows.
5. Set review dates and a route for reporting errors.
6. Ask a new starter to find the answers unaided. Only then consider search or an assistant on top.