Private AI for business, explained
A private AI system is an assistant that runs on infrastructure you control, connected to the tools your company already uses, that answers questions from your own material and shows its sources. For a company with several locations, it is also the one place where every location gets the same answer.
The phrase gets used loosely, so it is worth being precise. "Private" here means three specific things, and a system that misses any one of them is something else wearing the same word.
-
01
It runs on infrastructure you control.
Not a public chatbot, not a shared account, not a seat you rent by the month. The system sits on infrastructure your company controls, which is what makes the rest of the guarantees possible. Your data stays yours.
-
02
It is connected to your actual systems.
Drive, email, calendar, spreadsheets, project tool, CRM. A model with no connection to your material can only guess about your business, however capable it is. The connection is the product; the model is a component.
-
03
Every answer shows its sources.
An answer nobody can check is a rumour with better grammar. Each response names the documents and records it came from, so anyone can verify it in one click — which is the difference between a system people trust with real decisions and a novelty they stop opening after a fortnight.
What a private AI system does day to day
The whole team uses it through a plain chat window, in the words they already use. No query language, no training course.
- Answers questions that span several systems. "Which of our biggest customers has an open complaint and a renewal this quarter?" is three systems and a human today. It becomes one question with one answer.
- Settles which number is right. Where two systems disagree, rules agreed in advance decide which one is official — silently, on every answer, instead of a person arbitrating each time.
- Respects who is allowed to see what. Access follows the role. People see only what they could already see, and the system enforces that on every response.
- Watches instead of waiting. Tell it to flag a project running past its date or an invoice going unpaid, and it notifies the right person without being asked.
- Reads first. Every connection starts read-only. The system is only allowed to change anything once you have decided it should, for a specific recurring task, with clear permissions.
None of this requires your team to change how they work. That is deliberate: adoption fails when the system asks people to move, and succeeds when it meets them where the work already happens.
Private AI system vs chatbot licences vs a pilot
Three things get bought for this problem. Only one of them is a system.
- Chatbot licences for the team. Useful for drafting and thinking out loud. It has never read your contracts, so it cannot answer questions about your business or cite a source for what it says.
- An AI feature inside one tool. The assistant in your drive cannot tell you what the CRM says, and the one in your CRM has never seen the contract. Most questions worth asking need three systems.
- A pilot from an agency. It works in the demo, on last quarter's export. Six months later nobody owns it, the data has moved on, and the questions go back to the person who knows.
You have been buying intelligence. What you are short of is access.
There is a longer version of this argument, with the failure modes worked through properly, on the main page — and a direct comparison with a general chatbot in private AI vs ChatGPT.
How a private AI system gets built
Five steps. We call it the Company Brain System, and the order matters: people first, then the map, then security, then the brain, then automation.
-
01
Team training.
Your team learns to work with AI, with quick wins in their own daily work, before anything is connected.
-
02
Process, tool and org mapping.
Where information lives, where it gets stuck, and which tool is official for which number. This map decides what gets connected first.
-
03
Security review.
What data exists, who is allowed to see it, where it may live, and who signs off. Settled before more data is connected.
-
04
Your Company Brain.
Your processes, documents and tool data in one place, with the right access for every role. It starts small: three sources, not thirty, so the team uses it in the first weeks.
-
05
Automation.
Once the brain has earned trust, recurring work runs by itself, with clear permissions. Built and run for you, like an outsourced AI department, so no IT team is needed on your side. What an agent is, and the one security rule that matters, is in how to build AI agents.
Who private AI is worth it for
Worth it if
- You run an owner-led company with roughly 25–150 people.
- You have several locations, brands or companies.
- You have no in-house IT department.
- A growth step is coming in the next 12 months: new hires, a new location, a new company.
Not worth it if
- You have your own IT or technology team. You'd be better off building this in-house.
- You are just starting out and the knowledge still fits in one head.
- Everything already lives in a single tool that answers well enough.
Common questions — which tools it connects to, where the data lives, what happens if we part ways, what it costs — are answered in the FAQ.
Book a consultation call.
In 45 minutes, we look at where information gets stuck in your company, and you see the Company Brain live. If it doesn't fit, you still leave with two or three ideas you can use.
Book a consultation call