Company Knowledge Base: Build a Second Brain for Your Team
A company knowledge base is a second brain for a whole team: the answers people ask each other for, written once, owned by someone and readable by your AI tools. Most fail because nobody uses them. Here is how to start small, get it used and keep it right.
Founder of The Viable Edge, with 20+ years of experience helping businesses build competitive advantage through strategic transformation. Expert in AI-era business strategy and systematic implementation.
Published · Updated
To build a company knowledge base, start with the 20 questions your team asks each other most, write one short page per answer, and give every page an owner and a review date. Store it as plain text your AI tools can read, answer every repeat question with a link to it, and test it monthly with real questions.
What is a company knowledge base?
A company knowledge base is one shared place for the answers your team keeps asking each other: how you price, how a process runs, what was decided and why. Each answer is written once, owned by a named person and kept current. It is a second brain for a whole team rather than for one person.
A personal second brain holds what one person knows. A company knowledge base holds what the business knows, so it survives when someone is on holiday or leaves. We cover the definitions, the internal vs external split and the software options in what is an internal knowledge base. This post is about getting one built and used.
Why do most company knowledge bases fail?
Most fail on use, not on software. Someone builds a big structure, fills it once, and nobody owns the pages after launch. Within months a few answers are wrong, people catch it, and they go back to asking a colleague. A knowledge base that is trusted for 20 answers beats one that holds 500 nobody believes.
The pattern is the same in most small companies we work with. The launch is a project; the upkeep is nobody's job. So the build plan below spends more effort on owners and habits than on the first batch of pages, because that is where the failure happens.
AI raises the stakes. A person who reads a stale page might notice. An AI tool that reads it repeats it fluently to whoever asks, which is why a knowledge base for AI needs a source and a date on every claim.
How do you build a company knowledge base, step by step?
List the 20 questions people ask most. Write one short page per answer, with a source and an owner. Put the pages in one folder of plain text your AI tools can read. Answer every repeat question with a link instead of a reply. Then test it each month with real questions and fix whatever it gets wrong.
Collect the questions. Spend a week noting every question someone asks in chat, email or meetings. Pick the 20 that repeat. That list is your first table of contents.
Write one page per answer. Short, one topic each, with where the answer came from. Pull from the proposals, decks and docs people already trust rather than writing from scratch.
Name an owner and a review date. Usually the person closest to that work. A page without an owner is the first one to go stale.
Store it where AI can read it. Plain markdown in one shared folder works with every current AI tool. If your team lives in Notion or Confluence, check your AI tool can read it before you move anything.
Test before launch. Ask ten real questions, including two with no written answer, and score the replies. Our retrieval test has the scorecard.
The same five steps, at the scale of one person, are in our guide to building an AI second brain. Building your own first is the fastest way to learn what a team version needs.
Free download
Start with the Second Brain Starter Kit
The capture inbox, note template, source log, weekly checklist, connect-your-AI guide and retrieval self-test, as plain markdown files that work in Obsidian or any editor, with Claude or any AI. Enter your email and the download starts.
No spam, ever
Unsubscribe anytime
What should a company knowledge base include first?
Start with the answers that cost the most when they are wrong or slow: what you sell and for how much, how a new client is onboarded, who decides what, and the handful of processes everyone does weekly. New-hire questions are a good test. If a new person can get through week one from it, the base is useful.
Leave out what already has a better home. Contracts stay in the contract system and numbers stay in the finance tool; the knowledge base links to them and explains them. Meeting notes only belong once they hold a decision, and then as a short decision page with the reason, not the transcript.
For a marketing team, the first pages are usually the offer, the audience, the voice rules and the claims you can and cannot make. That brand slice is what we cover in what is a brand knowledge base.
How do you get your team to use a company knowledge base?
Make it the fastest way to get an answer. Reply to every repeat question with a link to the page, not a fresh explanation. Connect your AI tools to it so people can ask in plain language and get a sourced answer. When a page is missing, the person who answered writes it. Use follows speed, not mandates.
The link-instead-of-reply habit does two jobs. It sends people to the knowledge base, and it shows you which pages are missing, because every question without a link to give is a page to write.
AI changes the front door. Instead of browsing folders, people ask a question and get an answer that cites the page it came from. That only works when the pages are short, current and in a format the tool can read, which is why we compare tools by where they read knowledge from in AI knowledge management tools.
How do you know a company knowledge base is working?
Watch two numbers. First, how often the same question gets asked twice: it should fall. Second, a monthly retrieval test: ask the same ten real questions and score the answers. If scores drop, fix the page behind each wrong answer. A knowledge base that passes the test and cuts repeat questions is earning its upkeep.
Keep the test questions fixed for a quarter so the scores compare. Add two questions each month that the knowledge base should not be able to answer; a good one says it does not know instead of guessing. When that starts failing, a page is missing or an AI tool is filling gaps, and either way you know where to look.
For consultants and service providers
Build second brains for clients
Build and Deliver Professional Second Brains is a self-paced course on doing this for other people: scoping the engagement, building a knowledge base from a client's real material, testing retrieval, and handing it over, with a working knowledge base kit to build with. Plain files, no coding, no vector database.
