Ask a few people in your company what a customer's payment terms are, and you may get a different answer from each. The CRM says one thing. The accounting system says another. And the account manager remembers an email from last month that changed both.
Most businesses don't lack information. They have too many copies of it, in too many places, with no reliable way to tell which is current.
AI makes the problem harder to ignore. Statistics Canada reported that 19.2% of businesses used AI to produce goods or deliver services in the year before its spring 2026 survey, roughly triple the share two years earlier. An AI assistant that searches your shared drive will happily quote the first price list it finds, even an old one. Nothing tells it otherwise.
The fix owners ask us about most often is a company brain: one place where your business knowledge lives, stays current, and can be used by both people and AI tools. Here's what that means, how to start, and where these projects usually go wrong.
Where your knowledge actually lives
Take an ordinary distributor, contractor or accounting firm. Its working knowledge is spread across:
- Email, where many agreements with customers and suppliers are made.
- Shared drives, full of price lists, contracts and files named "final_v3_REAL".
- The CRM, which is only as current as the last person who updated it.
- The ERP or accounting system, which holds the official numbers but rarely the reasons behind them.
- Chat and meeting notes, where quick decisions get made and then scroll out of sight.
- People's heads, especially the long-serving staff who "just know" how things work.
Each is fine on its own. The trouble is between them. New hires don't know which source to trust, managers answer the same questions every day, and an AI tool connected to only some of them gives confident answers that are incomplete or out of date.
What a company brain is
A company brain isn't a new app everyone has to type into. It's a layer on top of the systems you already use, built from the parts below.
Connectors
Connectors pull information from the sources above. They run on a schedule or react to changes, so nobody copies anything by hand. The brain reads from your systems; it doesn't replace them.
An organized layer
Pouring every document into one search box isn't enough. A useful brain organizes what it pulls in around the things your business deals with: customers, products and prices, policies, suppliers and projects. When someone asks "what are this customer's payment terms?", the answer should come from that customer's record, not a keyword search across every file on the drive.
Freshness
This is the part most tools skip, and it matters most. Facts change, and the brain has to notice:
- A customer emails your accounts team: from next month, they'll pay on Net 45 instead of Net 30.
- The brain recognizes this as a change to that customer's payment terms that conflicts with what's on file.
- Instead of quietly overwriting the old value, it flags the conflict to the person who owns customer terms, with the email attached.
- That person approves the change, or rejects it if the new terms haven't been agreed.
- Net 45 becomes the current answer. Net 30 stays in the history, with the date it changed and who approved it.
Without that loop, a company brain slowly becomes another stale archive. With it, people can trust the answer and see why it changed.
Permissions
Not everyone should see everything. Salaries, HR files and margins belong to a small group. A company brain has to follow the same access rules as the systems it reads from, and apply them to AI agents as well as people. If a sales rep can't open the payroll folder, an agent working for that rep shouldn't quote from it either.
Answers with sources
Every answer should show where it came from (the email, the contract clause, the CRM record) and when it was last confirmed. People can check anything that looks off, and errors are easy to trace. An answer without a source is a guess, however well worded.
Start small
Begin with one team and one set of questions, not the whole company.
- Pick a team with a clear pain, such as customer service or sales support, who answer the same questions all day.
- Write down the questions they get asked again and again, and where they find each answer today.
- Connect only those sources. If the answers live in the CRM, one shared folder and the accounts mailbox, start there.
- Name an owner for each kind of fact: prices, customer terms, the returns policy. Owners approve changes.
- Run it alongside the current way of working for a few weeks, and compare.
Once that team trusts it, add the next one. Each step reuses the connectors and structure you've already built.
What to measure
Take a baseline before you start. The useful measures are simple:
- Accuracy: owners check a regular sample of answers.
- Sources: how often an answer comes with a source someone can open.
- Freshness lag: the time between a fact changing at the source and the brain giving the new answer.
- Open conflicts: how many flagged changes await approval, and for how long.
- Time to answer: how long staff take to find an answer, and how often they still ask a colleague.
On your own servers or in the cloud
On-premise keeps your data in your building and under your control, which suits firms with sensitive client files. The trade-off: someone has to run it, from updates and backups to hardware that can handle AI models.
Cloud is faster to start and leaves less to maintain. Check where your data is stored and processed, who can access it, and whether it can be used to train anyone's models. If it includes personal information, the Office of the Privacy Commissioner of Canada's guidelines say you stay accountable for it when a third party processes it for you, and that you should make it plain to people when it may be processed in another country.
Many businesses end up with a mix. We build on either; the right choice depends more on the data you hold and who will look after the system than on the technology.
Common pitfalls
Dumping everything in without ownership
Connecting every drive and mailbox on day one feels productive, but it fills the brain with old drafts, duplicates and contradictions that nobody owns. Every kind of fact needs an owner, or people won't trust the answers.
No freshness process
A brain that's accurate on launch day and never updated is worse than none, because people stop double-checking it. Catching and approving changes has to be part of normal work, not a yearly clean-up.
No permissions, or permissions added later
The first time an agent shows someone's salary or a confidential margin to the wrong person, the project loses the trust it depends on. Adding permissions afterwards is much harder than designing them in.
Treating it as a replacement for your systems
Your CRM, ERP and accounting system remain the official record. The brain reads from them, and approved changes should flow back to them. If it becomes a second place to edit data, you've created another copy to keep in sync.
Takeaway
You don't need a big project to start. This month, pick one team, list the questions they answer over and over, and write down where each current answer lives and who owns it. That list shows you where your knowledge is scattered, which sources to connect first, and who should approve changes. It's the first draft of your company brain, whatever tools you end up using.