BlogAI
AI for business: what it actually does
with your own data
Not the chatbot on your website. Four things artificial intelligence already does well inside a company, how to tell whether it pays off, and why it matters where the model runs.
The model is not the problem — your data is
A language model trained on the internet knows a great deal about the world and nothing about your company. It does not know what the 2024 supplier contract says, which product sells badly in March, or how your goods-in procedure works. That is why a generic AI subscription helps someone write an email faster but does not change how the company runs.
The difference comes from connecting it to your data. Same models, same technology, but with access to your documents, your stock and your history. That is where useful answers start.
Four things that already work well
Answers from the company documents
Contracts, quotes, procedures, manuals, spec sheets. You ask in plain language and get the answer plus the document it came from. A new hire becomes useful in days, not months.
Reading the documents that come in
Supplier invoices, delivery notes, orders arriving by email in ten different layouts. Fields are extracted and land in the management system without anyone retyping them.
Cameras that understand what they see
Not just recording. They recognise a person in a restricted area, a forklift on an escape route, a door left open. The alert arrives while it happens, not the next day when someone scrubs through footage.
Classifying and routing
Complaints, quote requests, tickets. They get read, categorised and delivered to the right person with a summary already written.
Where the model runs matters more than you think
There are two options and the choice is not technical, it is about data. In the cloud you use large models, pay per use and start within days. Your data does leave the company for an external provider, which for a medical practice, a law firm or a company with sensitive contracts can be unacceptable.
Locally, the model runs on a machine at your premises. Nothing goes out to the internet. The cost moves from a subscription into hardware, paid once, and speed depends on what you put in it. For a document assistant and for vision on cameras, one well-chosen server covers a small or medium company.
The simple rule: if the documents contain things you would not email to a stranger, the model stays in your building.
How to tell whether it pays off
An AI project that cannot be measured is a PR project. Before you start, write down which number has to move: minutes lost looking for a document, hours spent entering invoices, incidents caught on camera, time to first reply to a customer.
If that number cannot be measured today, the first step is not AI — it is starting to measure it. Otherwise you will never know whether you bought a result or a story.
- Pick one process, the one that hurts most
- Measure it for two weeks, exactly as it is now
- Start on a small slice, not on the whole company
- Compare the same numbers after a month of real use
Where to start, in practice
Most often, with documents. It meets the least resistance from people, because it does not change how they work — it just answers them faster. We index what you already have, without moving files and without rewriting anything, and permissions stay role-based: contracts are visible to the people who should see them.
Once that works and people trust the answers, it can extend to invoices, to cameras or to routing requests. In that order, not the other way round.
What this is about, in full
Frequently asked questions
How much does AI cost for a small company?
It depends where it runs. In the cloud you start with a small monthly cost proportional to usage. Locally, you pay once for the hardware and usage after that is free. For a document assistant in a small company, the local option usually pays for itself within a year.
Does our data reach the model provider?
Only if you choose the cloud option. When the model runs on a machine at your premises, nothing leaves the company. That is the only version you can guarantee in writing.
How long until we see a result?
On a small, well-chosen slice, two to four weeks until something is used daily. Projects that start across the whole company at once are the ones that never reach production.
Do we have to replace our existing software?
No. An AI assistant sits on top of what you already have: documents, stock, email. If your current system has a way to export data, it can be connected.