Skip to content
Etechan Butt

Home / Artificial intelligence for business

Applied artificial intelligence · companies

Reading invoices and contracts without typing a thing, answering calls, replying to the usual questions.

The artificial intelligence that works inside a company isn't a chat in a browser tab: it is a model put to work inside a specific process, with what can leave the company and what can't decided before we start. That is what I set up. And I teach your team to use it in their own work, not in general.

Book 15 minutes Request the X-ray
24 ha day of calls answered by an AI switchboard, at a law firm
96 %of invoices read and cross-checked with nobody touching them
Europeor your own infrastructure, for sensitive data
½ dayof training with up to eight people
What it really does

Four places where a language model pays for itself.

Not everything you can do with AI is worth it. These four are, and they are the ones I have measured inside companies.

01

Reading documents that arrive in any shape

Invoices, contracts, delivery notes, scanned reports. The model pulls out the data, puts it in the ERP's format and marks what it isn't sure about so a person can look. It doesn't matter that every supplier sends the PDF their own way.

02

Answering the phone when nobody is there

A voice switchboard that understands what the caller wants, handles the usual and passes to a person whatever needs judgement. At a law firm it answers around the clock and not a single call from a new client has been lost.

03

Replying to what always gets asked the same way

Opening hours, order status, what paperwork is needed, how long it takes. By email, by WhatsApp or from the website, with the answers you approve and not the ones it makes up.

04

An internal assistant that knows your documents

So the team can ask the procedure, the contract or the history instead of searching forty folders. With the documents inside your company, going nowhere.

The data

What leaves the company and what never does.

It is the first thing decided, before writing a single line. Some data can go to an AI provider in Europe under a data processing agreement, some can only be handled by a model installed on your own infrastructure, and some never goes into any AI system because it doesn't need to.

With health, children's or financial data, the default answer is inside. With a supplier invoice, outside but in Europe. And it gets written into the working session document, so management approves it knowing what it is approving.

I work with OpenAI, Anthropic and open models that get installed on a server of yours, and I choose according to the case, not the fashion.

8 peopleand half a day. That is what the training looks like: built on two of your processes and with the tools you already pay for.

AI training for teams, built on your own work.

No theory and no other people's examples. Half a day with up to eight people, two real processes from your company, and you leave with things set up and working. Plus something almost nobody teaches: which data can leave the company and which can't, so nobody pastes a contract into a public chat.

With Microsoft 365 Copilot, Google Workspace, ChatGPT or Claude, depending on what you subscribe to. If you have nothing yet, we decide in the session what is worth it.

Who it's for

Where it shows most.

Clinics and healthcare groups

Reminders people reply to, and handling the usual questions

With health data treated as a special category, inside the company.

See the case Distribution and import

Reading supplier invoices

Four hundred a month, as PDFs, each supplier doing it their own way. Read and cross-checked on their own.

See the case Construction and installations

Quotes that go out in hours

The model drafts it from the site visit and the technician only reviews it.

See the case Holiday rentals and non-hotel accommodation

Answering the guest at eleven at night

The usual gets solved on its own; whatever needs someone, alerts someone.

See the case
Questions

Before you book.

Does AI make mistakes?
Yes, and that is why it doesn't go where a mistake is expensive without a person behind it. It reads invoices and flags the ones it isn't sure about; it answers calls and hands someone the ones that need judgement. It is designed so the mistake lands in a tray, not in the accounts.
Does our data train someone else's model?
No. The work runs under processing agreements that forbid it, and for anything sensitive, on models installed on your own infrastructure where the data never leaves. What goes where is decided before we start and written down.
Is it useful for a small company?
If something repeats many times the same way, yes. A six-person clinic with two hundred appointments a week is a better case than a fifty-person company that does everything once.
Get started

Tell me what arrives in any shape and someone has to read.

Fifteen free minutes to assess whether AI is worth it in your case or whether ordinary automation is enough.

I need to know your name.
I need a number to confirm the time.
Check the email, it looks incomplete.
For example: «we receive about 300 invoices a month as PDFs and one person types them in». Tell me the case, even in one line.
When suits you

Tick as many as you like. I fix the exact time when I confirm.