The know-how stops belonging to one person
What the estimator with twenty years behind him knows stays in the company while he is on holiday. And it is still there the day he retires.
Home › Artificial intelligence
Almost everyone bolts artificial intelligence on top of the management system: it summarises, it suggests, it writes emails. We put it inside, where the data gets written. They are two different things, and only one of them saves any real time.
It is what we are asked for most often, and it is what changes people's day the most. The jargon calls it RAG. Here is what it does in practice.
Every company has an archive: contracts, specifications, data sheets, minutes, supplier manuals, correspondence, quotations from ten years ago. The answers to nearly every question you ask yourselves each day are somewhere in it. The trouble is that to find them you have to know where to look already — and the person who knows is always the same person.
We take that archive and make it answerable in your own words. Ask ‘what terms did we give this customer in 2022?’ or ‘which supplier guaranteed us delivery in five days?’, and you get the answer together with the document it came from. Not a plausible summary: the file, the page, the paragraph.
What the estimator with twenty years behind him knows stays in the company while he is on holiday. And it is still there the day he retires.
Each answer cites the document it was taken from. If there is no source, the system says it does not know — instead of inventing something that sounds right.
It runs on your servers or on our European cloud. Your contracts do not end up inside somebody else's model, and they are not used to train anything.
Not ‘generates some text’. It does the job: creates the customer, prepares the quote, books in the supplier invoice and loads three hundred items into stock. With the double-entry postings written behind it, the same checks and the same permissions as when you do it from the office. Set off by a voice note, from the company chat.
A voice note while you are out: ‘open a record for this new customer and put me together a quote from the price list’. By the time you get back, it is done.
A pallet turns up with the supplier's invoice. The document goes in, the items are booked, the stock levels move. Without going through the office.
It is not one more interface: these are the system's own operations, with the same checks and the same permissions. All that changes is where you set them off from.
This is how we work every day: our jobs, our quotes, our accounts. What we show you has been running on our own data for months.
The things somebody in the office redoes exactly the same way every day, and that add up to two hours by the end of it. We build an automation only when those hours genuinely exist and we have counted them.
Twenty emails a day: order confirmations, delivery notes, invoices, chasers. Somebody opens them one at a time, works out what each one is and passes it on. Now they arrive already sorted, with the attachment read and the figures inside the system. Whoever checks them reads, rather than retypes.
Every supplier sends its update in a different shape: one a PDF, one a spreadsheet, one an email typed out by hand. It used to be half a day of copy and paste every quarter. Now they come in by themselves and you get a list of what changed, to approve in ten minutes.
A vehicle inspection, an insurance policy, a contract renewing itself, a customer sixty days behind on payment. Nobody has the time to check every day. The system watches instead of you and tells you in the chat, before it becomes a problem.
What you invoiced, what you collected, which jobs are losing margin, what is missing from the stores. Somebody used to put it together by cross-reading three screens. Now it turns up written, with the figures taken from the system instead of copied out.
There is no other application to learn, and no twenty tabs open. You ask the system from the same place you already talk to your colleagues.
Built on Matrix, on your servers. Type or send a voice note in the right group, and the request lands where it should.
The result arrives in the same conversation, with the document attached. If something does not look right, the colleagues reading can see it too.
Every operation carries the name of whoever asked for it, the time, and the message that set it off. It is not a black box.
Before we put anything to a customer we keep it running on ourselves for months. These are the two tools we run the company with — and almost everything we later bring you comes out of them, already proven.
What is open, who is working on it, what falls due this week. Plus the automations that read the post, keep an eye on the deadlines and put the summaries together: the same ones that, once they have run long enough, end up with customers.
Every decision taken stays written down with its date and its reason, hooked to the work that produced it. It is why, years later, we still know why a thing is built the way it is — and do not break it trying to improve it.
Because it is the only proof worth anything: a supplier who does not use what he sells has never watched it break. Our working days have run inside these two for years, on real data: ours.
Aura and Atlas are our own tools, not two products to sell you. What we bring to you is the method they taught us, not our desk.
Over the past few years Italian firms have moved conversations, documents and customer records into services they do not control, often without noticing. We work the other way round: first you settle where the data sits, then you choose the tool.
We use models we can run on our servers or on yours. Your contracts and your customer records do not go into anybody else's model, and they are not used to train anything.
When an automated system writes a record, it stays on file that the system did it, when, and at whose request. It is what the European regulation on artificial intelligence asks you to be able to demonstrate — and it is worth having anyway, obligation or not.
Data in Europe, access logged, backups tested. No transfer outside the Union tucked away in a clause, because we do not use services that provide for one.
Quotes, site photographs, customer details, decisions taken at eleven at night: all of it passes through a Meta service, on personal phones, with no way for the company to export it, search it or delete it. When a person leaves, the conversation leaves with them.
Built on Matrix, the open standard for messaging. It runs on your own infrastructure: the messages are yours, not lodged with somebody else.
Groups, files, end-to-end encryption, voice and video calls, apps for the phone and for the desktop.
Someone who leaves the company loses access, and the conversation stays behind. On WhatsApp it works the other way round.
The same chat you coordinate the work in is the one you ask the system to prepare a quote from. One channel.
We are not legal advisers and we do not sell certifications. We build systems that make these things demonstrable: who has access, where the data sits, what an automation did and when. The rest you check with your own adviser, using the documents we put in your hands.
A searchable archive is designed around the real documents: which ones there are, where they sit, what state they are in. The first step is always the operational review — then you decide what is worth building.