How to build your own company brain: lessons from our own attempt

One of the highest-value things a business can build with AI is what we call a "company brain": an assistant that actually knows how your organisation works — your processes, your decisions, your data — and can answer questions about it. We built one for ourselves. It was genuinely useful, and it taught us more than we expected. Here's what we learned, including the parts we'd do differently.

Why a company brain beats a generic assistant

A general-purpose AI is a brilliant, well-read stranger. It can help with almost anything in the abstract — but it knows nothing about your pricing logic, your onboarding process, or why you dropped a supplier last year. A company brain closes that gap by grounding the model in your own trusted sources. The technique behind it (retrieval — fetching relevant internal information and answering from it) matters less than the shift it creates: from "smart stranger" to "colleague who's read everything."

Lesson 1: your knowledge is messier than you think

The biggest surprise wasn't technical. It was discovering how scattered and inconsistent our own knowledge was — the same policy described three different ways, decisions no one had written down, documents that contradicted each other. An AI grounded in that mess faithfully reproduces the mess. Most of the real effort went into getting our knowledge into shape, and that work paid off on its own, AI or not.

Lesson 2: start from questions, not documents

Our instinct was to feed it everything and hope. That's backwards. What worked was starting from the questions people actually asked most often — "what's our policy on X?", "how do we usually price Y?" — and making sure the brain answered those few things brilliantly. A tool that nails the top ten real questions beats one that vaguely covers a thousand hypothetical ones.

Lesson 3: trust is earned, not switched on

The first time the brain gave a confident wrong answer, people stopped trusting it — fast. We learned to design for that: show the sources behind every answer so people can verify, let it say "I don't know" instead of guessing, and roll it out on low-stakes questions first. Trust compounds slowly and collapses instantly; build for the slow version.

Lesson 4: it's a living thing, not a launch

A company brain isn't something you switch on and walk away from. Your business changes, and the moment the brain's knowledge goes stale, its answers quietly become wrong — which is worse than no answer at all. Someone has to own keeping it current. Budget for that from the start, or the thing decays into a confident liability.

Was it worth it?

Yes — but not only for the reasons we expected. The assistant itself saves real time. The deeper payoff was being forced to understand, organise, and write down how our own business actually works. That's an asset in its own right, and most organisations never do it until something makes them. If you're thinking about building one, that's the honest pitch: you'll get a useful tool, and you'll get a much clearer company along the way.