What Is a Data Brain? The AI That Runs Your Business Data

What is a Data Brain?

A Data Brain is a private AI that knows your business. It reads your own data — sales, stock, customers, emails, files — and lets you talk to it like a colleague. You ask a question by voice or text, and it answers from your real numbers.

Think of it as one assistant sitting on top of every system you already use. Your shop, your spreadsheets, your inbox and your accounts stop being separate islands. You stop opening five tabs to answer one simple question.

Three things make it a Data Brain rather than just another chatbot:

• It knows. It is connected to your actual business data, not the public internet.
• It thinks. It spots what changed, what is late and what looks wrong — and tells you first.
• It does. Once you allow it, it handles routine jobs like reminders and reports, with risky actions sent to you for approval.

How is AI changing the way businesses use their own data?

The shift is simple: you no longer need to know where data lives to use it. For years, getting an answer meant building a report, exporting a sheet or asking the one person who understands the system. AI turns that into a plain question.

Most small and mid-sized businesses already have plenty of data. The problem is that it sits in different places, in different formats, and nobody has time to join it up. So decisions get made on gut feeling, or on last month’s numbers.

Language models changed that in two ways:

• They read messy data. Emails, notes, PDFs and half-filled spreadsheets are no longer a blocker. The AI can make sense of them.
• They answer in plain words. You ask “who hasn’t paid this month?” instead of learning a reporting tool.

The result is that the owner — not just an analyst — can talk to the business directly. That is the whole idea behind a Data Brain.

How is a Data Brain different from ChatGPT, a dashboard or an AI agent?

The short version: ChatGPT knows the world, a dashboard shows numbers, an agent does one job — a Data Brain knows your business and talks about all of it. They are not rivals. They solve different problems.

In practice they work together. Many businesses start with an AI agent for customer replies, then add a Data Brain so the owner can see and question everything. If you want the bigger picture on agents, read Agentic AI explained.

What can you actually ask a Data Brain?

Anything you would normally ask a manager who has every report open. The best questions are the ones you ask every week and currently answer by hand.

A few real-world examples, by type of business:

• Online store: “How did we do today compared to a normal Tuesday?” “Which product is about to run out?” “Who ordered twice but never came back?”
• Service business: “Which quotes are still waiting for a reply?” “Who hasn’t paid their invoice?” “What did we earn from repeat clients this month?”
• Several websites: “Any orders across all sites today?” “Did that customer’s email actually arrive?” “Is any site slow right now?”
• Back office: “Where is the signed contract for this client?” “What did we agree on price last time?”

The follow-up is where it gets useful. After “three invoices are unpaid”, you can say “send them a polite reminder” — and, if you have allowed it, the brain does it and reports back.

How do we build a Data Brain?

In four stages, and it starts by answering questions before it is allowed to act. A first version is usually live in days, not months.

1. Map your data. We list where your business lives — store, sheets, inbox, accounts, CRM — and the first ten questions you want answered.
2. Connect and organise. Each source is connected securely with its own access keys. Duplicates are cleaned up and everything goes into one private memory the brain reads from.
3. Write the rules. What it can answer, what it can do alone, and what always needs your OK. These rules are written in plain words and agreed with you.
4. Talk to it and grow. You start asking. Each week we add the next question it should handle and, when you are ready, the next job it can take on.

Step one matters most. A brain that answers the ten questions you actually ask is worth far more than one that can, in theory, answer anything. If your data is spread thin, a custom web app that pulls it into one place is often the first piece of work.

What did we learn running our own Data Brain?

We built one for ourselves before building one for anyone else. Our own Data Brain watches 13 websites and 2 servers — orders, payments, emails, errors and visitors — and we ask it questions out loud.

Four lessons came out of that, and we now build every client brain around them:

• Start read-only. Ours only reads and reports, by our choice. It tells us what is happening; we decide what to do. Trust comes first, actions come later.
• Let it rest. It only works while someone is talking to it, and sleeps otherwise. That keeps the websites fast and the running cost close to zero on quiet days.
• Short answers win. “Any orders today?” should get a number in seconds, not a paragraph. We tune for that.
• Voice is a luxury, text is the workhorse. Talking to it is brilliant when your hands are busy. For everyday checks, typing is faster to scan and far cheaper to run.

What does a Data Brain cost to run: text or voice?

Typed questions cost very little. Live voice costs many times more. The build is a one-time project; the running cost depends on how much you use it and how.

At the time of writing, Google’s published Gemini API prices give a clear picture:

• Text: a light model such as Gemini Flash-Lite is priced at $0.30 per million input tokens and $2.50 per million output tokens. A typical business question with its data comes to a small fraction of a cent.
• Live voice: Gemini’s live audio models are priced per minute — about $0.005 for each minute it listens and $0.018 for each minute it speaks — before the cost of re-reading a long conversation.

What that means in practice: in our own testing, a few minutes of live voice cost more than a full day of typed questions. So we set clients up with text for daily use and voice as an option, with a daily spending cap so a long chat can never surprise you.

The build itself is quoted per business, because it depends on how many systems are connected and what the brain is allowed to do. Tell us your questions and you get a fixed written quote.

Is your business data safe inside a Data Brain?

It can be — if it is built like a bank vault, not a chat window. A brain that knows your whole business is valuable to you and to anyone who wants in. Security has to be designed in from day one.

Two public standards are worth knowing:

• The NIST AI Risk Management Framework, a voluntary guide built around four functions: govern, map, measure and manage AI risk.
• The OWASP Top 10 for LLM applications, which lists risks such as prompt injection, sensitive information disclosure and “excessive agency” — an AI allowed to do more than it should.

In plain terms, every brain we build follows four rules:

• Least access. It connects only to what it needs, with its own keys you can switch off at any time.
• Your OK for risky actions. Payments, refunds and deletions always come to you. That is the direct answer to excessive agency.
• Private model use. We use paid AI services whose terms keep your data out of product training. Google’s Gemini billing page, for example, states that on the paid tier your prompts and responses are not used to improve Google products.
• Everything logged. Every question, answer and action has a record you can read.

When do you not need a Data Brain yet?

When your answers already live in one place. A Data Brain pays off when information is spread across several systems. If that is not you yet, start smaller and save the money.

• Your business runs on one spreadsheet. A simple report or a single automation will do the job.
• You need one task done, not answers. If the goal is replying to customers or booking calls, an AI agent is simpler and faster.
• Nobody will ask it anything. If you cannot name five questions you would ask it every week, wait until you can.

Messy data is not a reason to wait. That is normal. Cleaning and connecting it is part of the build — just expect it to take most of the first week.

How to get ready for a Data Brain this week

You can do most of the groundwork in an afternoon. These five steps make the build faster and the quote more accurate.

1. Write down ten questions you answer by hand every week. These become the brain’s first job.
2. List every system your business uses — store, sheets, email, accounting, booking, CRM.
3. Mark the sensitive ones. Payroll, personal data, bank access. Decide what the brain should never touch.
4. Pick text or voice for daily use. Text is the sensible default; voice is a nice extra.
5. Decide your first action. One routine job you would happily hand over once you trust it, such as payment reminders.

Send that list through our Support form and we will reply within one business day with a plan and a fixed quote. You can also see how the service works on the Data Brain page.

Frequently asked questions

Sources

• Gemini Developer API pricing — Google
• Gemini API billing — Google
• NIST Risk Management Framework Aims to Improve Trustworthiness of Artificial Intelligence — NIST
• OWASP Top 10 for LLM Applications — OWASP GenAI Security Project