Artificial intelligence applied to specific services — not a general‑purpose assistant.

The problem we open with

Companies today have capable tools — and they use them as isolated toys.

• Information comes in through ten channels email, messages, documents, phone calls — and nobody collects it in one place. The person who has to decide walks into the meeting without the picture.
• Repetitive work stays on people’s desks. Preparing a meeting, routing a request, updating a document: hours of skilled work spent on work that does not qualify anyone.
• General-purpose tools do not know the business. They answer well in the abstract and badly in your context, because they do not have your method, your documents, your history.
• Nobody knows what an agent actually does who authorised it, what it can reach, where the data goes, who can stop it. So trust becomes a gamble, and adoption stops there.

The problem is not a lack of capability. It is a lack of structure — and of a process worth improving.

How the system works, in nine steps

Information enters the way it already enters. What changes is where it ends up — and what happens next.

1. It comes in. Information arrives the way it already arrives — email, message, document, phone call. Nobody has to change how they work: what changes is where it ends up.
2. It is collected. Everything that comes in lands in one place, in order, with its origin. The first gain is that the picture exists.
3. It is interpreted. The system recognises what this is — who is writing, what they are asking, which process it belongs to, how urgent it is — and decides whether and where to act.
4. It retrieves the company’s material. The real knowledge: documents, previous cases, the rules of your method. The intelligent system is not the archive — your data stays where it lives, and the system connects it. When the material is not enough, it says so.
5. It prepares. The draft of the action — a reply, the preparation for an appointment, a document, the summary you decide on — with the sources cited.
6. It decides who decides. Low risk → it acts on its own (spot checks) · medium → it proposes, an approval is required · high → a person decides, always.
7. It acts and updates. The action is carried out and the systems are updated accordingly. No re-typing by hand, no step where information gets lost or changes shape.
8. It leaves a trace. Every step is recorded: what arrived, what was read, what was proposed, who approved, what went out and when. The trace lives out of reach of the system that acts.
9. It measures and improves. What it collected, what it handled alone, what it brought to a person, what was corrected — and a rhythm for reviewing and correcting the process.

Where we start

Many opportunities — start from one to three processes with the highest value. Not ten at once.

The five rules you will see in every project

1. The method is yours. No system changes a process, a rule or an approved document without your go-ahead. Human validation is a blocking gate, not advice.
2. We talk about capability, never about tools. In the documents you read you will not find the name of a supplier, a model or a technology: you will find what the system can do.
3. Your data stays where you decide. Where the material lives is decided together, deployment by deployment, and recorded in the project.
4. Nothing is deleted behind your back. The normal behaviour is to archive; deletion happens only when requested or authorised, and it leaves a trace.
5. You see it before they receive it. Every message that goes outside passes through you.