The highest ROI AI project for a business is a company brain

  • company brain
  • getting started with AI
  • AI strategy
  • Claude

If you ask me where your business should start with AI, my answer is the same every time. Build a company brain first. Of everything I have built, it is the project that pays back fastest, because every other thing you do with AI gets better once it exists.

I say this to friends who run companies and I say it in every AI talk I give. Here is what I mean, what it looks like at my own company, and how to start one.

What a company brain is

A company brain is one shared memory of your company's calls, email, chat and decisions, and every AI tool your team uses reads it before it starts work. Each memory is a short fact with a link back to where it came from. Who promised what to which customer. What you decided about pricing in March. Why you stopped selling a package last year.

It does not replace your CRM, your help desk or your shared drive. Those stay where they are, and they stay the place of record. The brain holds the fact and points to the source.

Why it beats every other first project

Most companies start with one task. Write my emails. Summarize this document. Answer support questions. Each of those works a little, then stalls, because the AI knows nothing about your company. Someone pastes in the background every time. When that person is out, the quality drops.

A brain fixes the thing under all of those tasks. Once it exists, the support helper knows what the sales team promised. The follow-up email knows what was said on the call. A new hire can ask "what did we decide about the Acme renewal?" and get the answer with a link, instead of interrupting three people.

That is why the return compounds. You build it once, and every AI project after it starts with your company's context already loaded. Pick any other first project and you rebuild that context by hand for each one.

Corrections also stick. When you fix something in a normal AI chat, the fix is gone when the chat ends. With a brain and a short page of written rules, a correction made on Monday still holds on Friday, for every person on the team.

What it looks like at my company

I run Prentus, an AI company in higher education. Before we had a brain, every AI chat started from zero. I explained the same background to every tool and every teammate, over and over.

Today our brain holds about 38,000 memories, about 17,000 of them active, and all of them are searchable. They come from email, Slack, our calendar, our CRM and our calls. Every sales call is read by an AI that logs the commitments, decisions and blockers. From October 14 to November 6 that was 23 calls and 366 facts, saved without anyone taking notes by hand.

Every AI helper we run reads the brain before it starts work. That includes the morning sales review, which finds every follow-up I owe and drafts it in my inbox. Nothing sends until I say so.

It was not clean on the first try. Early on, the same meetings were processed again and again, and rephrased copies of the same facts piled up. When we caught it, we fixed the cause in one day and archived 21,247 duplicate rows. Now every new fact is checked against what is already there before it is saved. I share that because your brain will have bugs too. A weekly review catches them early.

How to start one

You do not need a data team or a big software project. The order matters more than the tools.

  1. Write your rules down. One page. How you always work, like "never send an email to a customer without a person approving it." Every person's AI follows the same page.
  2. Save decisions on purpose. Ask each team to log ten real decisions. Ten decisions per team beat a wiki nobody finishes.
  3. Connect your call recorder first. Calls give the most value per hour of setup, because that is where customers say what they want and where decisions get made.
  4. Add email and team chat next. Email can be read live through a connector, so nothing gets copied. For chat, start with shared channels only.
  5. Name one owner. One person spends 30 minutes a week skimming new memories, archiving noise and turning repeated corrections into rules.

The technical setup is a small database and one connector that lets Claude search and save. I use Claude Code for that part. Claude Code is an app from Anthropic, the company behind Claude, that works on your computer's files and tools for you. You describe what you want in plain English and it does the setup. No coding needed.

The install takes about one working day for the person doing it. Getting people to use it takes about four weeks. Start with 3 to 5 people who use AI every day and will tell you what is broken. Fix what they hit, then roll it out to everyone.

What about private information

This is the first question every owner asks me, and it should be. A few rules cover most of it.

  • Store short facts and links, not raw messages or whole documents.
  • Give every memory a visibility level, so private notes never show up in a team search.
  • Keep sensitive calls private by default.
  • Give people a word that keeps a personal chat out of the brain.
  • Keep passwords and keys out of it completely.

We learned some of these the hard way. A personal chat about weekend plans once reached our team brain. That is why the opt-out rule exists now.

The mistake to avoid

The most common wrong start I see is a big "get all our knowledge down first" project. Someone opens a wiki, writes three pages, and it dies. The other wrong start is importing every document and email into a new database. You end up paying to store a copy of everything and the AI still cannot tell what matters.

A brain works because it holds facts and decisions, small and linked, and it fills itself from the tools you already use.

If you want my exact setup, it is in Your AI Starter Kit for $99. It has the rules file, five ready-made helpers, the brain and a four week plan, and it works with Google or Microsoft. If you would rather have me alongside while your team builds it, I also run a month of four short sessions that ends with Claude working from your company's rules and memory, all in your own accounts.

Not sure your company is ready for a brain yet? Take the free 2 minute quiz. Nine questions, and you get a score and the order I would fix things in.

Questions

How long does it take to set up a company brain?

The install takes about one working day for the person doing it. Getting your team to use it takes about four weeks, starting with 3 to 5 people and then the whole company.

Is a company brain safe for private information?

It can be, if you build in privacy from day one. Store short facts and links instead of raw messages, give every memory a visibility level, keep sensitive calls private and give people a way to opt out of personal chats.

Do I need a developer to build one?

No. You need one owner who can follow written steps. Claude Code writes and sets up the technical parts for you from a plain English description.

Was this useful? Score your AI readiness

Nine questions, 2 minutes, free. You get a score out of 100 and the order I would fix things in.

Take the 2 minute quiz