
How Do I Know If My Company Brain Is Working?
Test your company brain with twenty real customer questions. Check each answer is right and points to its source, then fix the document, not the prompt.
Part of AI for local service businesses
Test it with twenty real questions. Collect twenty questions customers or your team actually asked last month, write down the right answers, then ask your AI agents the same questions while they're connected to your company brain. Check two things for each answer: is it right, and does it point to the document it came from? When an answer is wrong, fix the document, not the prompt.
Why twenty real questions?
Because real questions are the ones that cost you money when the answer is wrong. Made-up test questions tend to be easy. Real ones are messy, specific and asked the way customers actually talk.
Twenty is enough to cover your prices, your policies and the questions that come in every week, and few enough to run in an afternoon.
Where do I find them?
Your inbox, your DMs, your booking notes and your front desk. For a med spa, last month's twenty might include:
- How much is lip filler?
- Is the deposit refundable if I cancel?
- How long does Botox last?
- Can I work out after a HydraFacial?
- Do you have Saturday appointments?
- Do you offer payment plans?
Add a few your team asks, too, like "What's our cancellation fee?" or "Who handles a complaint?"
How do I write the right answers?
Short, and from the business: the answer you'd want your best person to give. Next to each one, note which document it should come from. If there's no document for it, you've found your first gap.
How do I run the test?
- Connect the agent to your company brain.
- Ask each question in a fresh chat, the way a customer would ask it.
- Mark each answer right, wrong, or right but with no source.
- Note which document it used.
Run it on every AI tool that answers for your business. If two tools give two different answers to the same question, the fact lives in two places. Here's why that happens.
What do I do with a wrong answer?
Fix the document, not the prompt. A prompt fix helps one agent in one tool. A document fix helps every agent that reads it.
A wrong answer is almost always one of three things: the fact is missing, it lives in two places that disagree, or it's out of date.
Why does the source matter?
Because an answer you can't trace is an answer you can't fix. When an agent points to the document it used, you know exactly where to look when something's off, and you can check it before it reaches a customer.
How often should I rerun it?
Keep the twenty questions, and rerun them whenever the brain changes: a new price, a new service, a new agent. Add new questions as customers ask them. Over time, the test becomes a record of what your business knows. Here's how to keep a company brain up to date between tests.
If you haven't built yours yet, start with the five documents I'd put in first, and part two of my series walks through the full build.
What do I use?
Ownbox is the anti-SaaS Business OS you actually own. Your team, your AI agents, your company brain and your Machines work together on a server your business owns, and after 12 payments the software is yours too.
This is how I check my own Ownbox. Claude, ChatGPT and the agents working inside it all read the same AI Brain, and when my Ownbox drafts a reply, it shows the documents it used right under the draft. When one answer is wrong, I fix one document, and every agent gets it right.
Everyone is building AI agents. We built a place for them to work, together.
You don't know your company brain works until you've asked it what your customers ask.
