What is a company brain? And what it isn't

A company brain is the place your AI reads before it answers on behalf of your business. It holds how your company actually works: who gets a refund and when, which discounts sales may give, how support phrases a delay. Every rule in it has someone who stands behind it.

The term is everywhere right now, and it gets used for very different things. Some vendors mean enterprise search across all your tools. Some mean a personal note-taking app with a chat window. So here’s how we use it, and why the difference matters once real customers are on the other end.

The problem it solves

Most companies we talk to already use AI. Usually more than one: ChatGPT in sales, Copilot in the back office, Claude for someone who writes a lot. Each tool knows a little about the company. Somebody pasted the price list into one chat, somebody else wrote a long custom instruction in another.

That works until two people ask the same question and get different answers. One assistant says the return window is 30 days, another says 14. Neither is lying. They were each told something different, months apart, by different people.

The knowledge isn’t missing. It’s scattered, and nobody decided which version counts.

What a company brain is

We keep the definition narrow on purpose. A company brain has four properties:

1. It holds rules, not documents. A rule is a short statement with a reason: “Returns are accepted within 30 days because our supplier takes them back within 45.” Rules are small enough to check, change and approve one at a time.
2. Every rule has an Owner. The head of customer service owns returns. The CFO owns payment terms. Nothing counts until its Owner has approved it.
3. Contradictions surface. When two rules disagree, both Owners see it. Neither wins by default.
4. Every AI reads the same version. ChatGPT, Claude, Copilot and your own tools connect to one place, today usually over MCP (Model Context Protocol), and read only approved rules. When an answer looks wrong, you can trace it to the rule and fix it once.

If any of these is missing, you have something useful, but it isn’t a single source of truth for your AI.

What it isn’t

A wiki. A wiki stores pages and hopes someone reads them. Pages go stale quietly, and an AI that reads a stale page repeats it with full confidence. A wiki can be a good starting point for a company brain, though. Most rules are written down somewhere already.

A second brain. A personal second brain collects everything you might need someday: notes, links, highlights. That’s great for one person. For a company, “everything” is the problem. The AI can’t tell the draft from the decision. A company brain holds less, on purpose: only what someone approved.

Enterprise search. Search tools index your drives, chats and tickets so you can find things. That helps people. It doesn’t decide which of five conflicting documents is right, and that’s what the AI needs before it answers a customer.

A bigger prompt. Custom instructions and project files in ChatGPT or Claude are a fine first step. They live inside one tool, though, they have no owner, and nobody notices when they go out of date.

How the knowledge gets in

This is the part most teams underestimate. The rules that matter are rarely in a document. They’re in the heads of the people who’ve done the job for years: the exception for that one wholesale customer, the reason invoices above a certain amount go to the CFO.

We collect them in three ways:

• A short sketch. The manager names the areas and who owns each one. Fifteen minutes, by voice.
• What already exists. Manuals, checklists, old wiki pages, imported as a starting point, not as the truth.
• Interviews. The AI talks to the people who know: the normal case, the exceptions, two real examples. It asks about gaps and contradictions in what’s already there.

From that come draft rules. Owners approve them, and only then does any AI see them.

When you need one

Not every company does. If five people share one AI tool and talk to each other every day, a well-kept project file may be enough.

It starts to pay off when:

• several teams use AI and the answers drift apart,
• AI writes to customers in your name (you own what it says, as Air Canada learned in court in 2024),
• a few experienced people carry knowledge nobody else has written down,
• you want to switch or add AI tools without rebuilding your setup each time.

How to start without buying anything

Pick one thing your company does often, returns or quotes or onboarding. Write one page: the steps, the rules, what good looks like, two real examples. Name the person who owns it. Give the page to the AI you already use.

That page is a company brain with one rule set in it. Once you have five of those pages and three AI tools, you’ll feel why approval, conflict checks and one shared source start to matter.

That’s what we build with CtxCore Brain: the interviews, the Owner approvals and the delivery to every AI tool, set up with you and kept current.