Tailored AI systems
for Italian SMEs.

The request that arrived on Saturday morning, answered by someone on Monday at nine.

They come from WhatsApp, the website, email, the phone. Whoever answers first gets the order.

The quote sent and never picked up again. The customer who bought only once.

Nobody said no. That contact was simply never given a next step.

The March invoice, collected in May.

Overdue invoices get chased when someone remembers, and someone remembers when the day allows it.

Data copied from one sheet to another, with the occasional error that slips through and is found later.

A machine would do it the same way, and it doesn't skip a line when the phone rings.

They search for exactly what you do, and open someone else's website.

It happens on a search engine as well as inside an AI assistant, which by now answers in place of the list of results.

The renewal discovered the day after, the grant call read after it closed.

Nobody checks them because they are nobody's job.

A hundred real requests: orders, reorders, complaints, customs questions, reminders. Four channels, five languages.

A hundred real requests across four channels and in five languages: these are the seconds from arrival to a draft ready for approval, median over a hundred. Twenty arriving in the same second were all closed, all twenty, in twenty-one seconds, which is the moment when a person has to choose which one to ignore. A hundred in, a hundred out with their reply, and the right priority assigned a hundred times out of a hundred: complaints and customs at the top, payments and junk mail at the bottom.

By hand the same request takes four to ten minutes, and that is a declared estimate: read, understand, look up the data, write, reread. The seconds are timed, the minutes are not.

A contact writing on WhatsApp, read and queued with their next step already assigned.

Median over nineteen contacts that actually arrived, not simulated. Nobody is left without a next step, which is the point: contacts don't go cold by themselves, they go cold when nobody touches them.

Thirty working documents: invoices, delivery notes, technical sheets, orders. Five languages, three different layouts.

Two hundred and thirty-four correct fields out of two hundred and thirty-seven, and the document type recognised thirty times out of thirty: invoice, delivery note, technical sheet or order, in five languages and three different layouts. We had hidden an unreadable one in the pile on purpose, and it set it aside instead of making up the numbers.

The three missed fields have a reason, and we state it: twice the same defect, an amount equal to zero left blank instead of written as zero, and once a country that wasn't written on the document and had to be inferred. None of the three is an invented value.

Open the PDF, find the fields, copy them into the management software, check again.

Six point three seconds per document, median over thirty. By hand the same document takes three to eight minutes, and that is a declared estimate: the seconds are timed, the minutes are not.

The founder of a management SaaS for the beauty sector was looking for Italian professionals to work with, within a specific window of time. Five professions, six cities, no bought list.

Twenty-two searches, three hundred and eighty-nine names collected, ten delivered in order of priority: 2.6%. In between, two filters, the thresholds he set himself, and a manual reading, profile by profile, of the thirty that remained. One hour and four minutes from the first search to delivery. By hand the same work takes a week, and somewhere around the two-hundredth name you stop reading and start scrolling.

For three of the ten, the problem had been written in public by them: one had asked customers to write to him again because he had lost some messages, two take bookings by hand. For those three the first message was already ready, built on what they had published.

We measured the other rows on this page on our own systems. This one was received by a person outside Itria, and the sentence is theirs, written in public: “Within an hour I received a PDF with all the information.”

Five public exhibitor lists, no contact bought, no paid list.

Twenty minutes in all, for what takes a salesperson a whole day. Four hundred and twenty-eight names read, sixty-four websites found and confirmed one by one, ten emails written with a defect verified on that single company's website: none of the ten could be sent to anyone else.

A tube-bender manufacturer, a restaurant, a domain for sale, a Chinese gaming page.

Nineteen companies had the right name and someone else's website. They were stopped before entering the list, not after the email went out: the step that caught them is checking the website against its content, and it is the one that pays off most in the whole chain.

An external check that asks the machine whether the systems are running, every five minutes, day and night.

An external check asks the machine whether the systems are running every five minutes, day and night, for a hundred and twenty-four days: about thirty-five thousand seven hundred checks. It comes from outside, so it doesn't depend on the machine it is watching.

And underneath, every single process is logged: when it started, what it worked on, how it ended. Three hundred and fifty-six log entries between 18 May and 26 August. No engine runs in the dark, and when something breaks the response is human.