The company
Omnia Technologies is an industrial group based in Trevignano, in the province of Treviso. It brings together Italian companies that build machinery and systems for wine, beverages and food, and for life sciences: pharmaceutical, nutraceutical and diagnostics.
Its story starts in 2020. Since then it has grown through acquisitions, more than one every year, and in 2023 it took the name Omnia Technologies. Every company that joined the group brought its own machines, its own manuals and its own way of working.
Alongside the machines there is service: support, spare parts, upgrades and training for the whole life of a plant. When a machine stops at a customer’s site, it is customer service that has to find the fix.
The group started working with Yellow Tech in February 2026, with an AI training program for its people. The agents came right after, starting from the processes the group had already mapped.
The challenge
Customer service works clients’ cases in the company’s CRM. Fixing a fault takes the right procedure for that machine family and that model, and the procedure lives in different places: PDF manuals on different servers, support emails, messages between technicians, cases already closed.
On top of that there was an archive of intervention reports, filled in by field technicians and stored outside the CRM. Opening a customer’s case, the history of the interventions on that machine was not visible.
Searching all of that by hand makes every case longer. The fix for a fault seen before often lives in a senior technician’s head: newer staff have to ask them, and when a senior technician leaves the company, that knowledge leaves too.
The solution
Yellow Tech built an AI agent for customer service. The operator opens the case in the CRM, opens the agent’s panel and asks how to solve it. The agent reads the case context on its own (product family, machine model, error code, fault description), so the operator doesn’t retype anything.
The agent searches by meaning across the technical manuals, the articles of the knowledge base and the cases already solved, and proposes the procedure. Every answer cites its sources: one click opens the manual or the case the fix comes from. At the end the operator rates the answer (solves it, useful but incomplete, off track), and that rating is used to improve the system.
The work went in three phases. First the manuals: the documents in the shared archive were sifted, labeled by sector and by machine, and loaded. Then the history: historical support emails and the intervention reports became articles of the base, and every report became a historical case in the CRM with date, customer and machine. Finally the automations that keep the base up to date every day.
Articles start as drafts, and an expert at Omnia validates them before they enter the base. The drafts come from new support emails, from the closing of every case in the CRM and from the end-of-intervention reports that technicians fill in on site.
The panel sits inside the CRM, while the knowledge base and the search engine run on the company’s cloud. That way the same base can also be queried from other channels.
The results
The project for Service started on May 26, 2026. The agent has been released and is in use in the technicians’ console.
Operators no longer hunt for the procedure across folders and mailboxes: they ask from the open case and get it with its sources. The history of the interventions on a machine, which used to sit outside the CRM, can be read in the historical cases. And every closed case or end-of-intervention report becomes a draft article, so what technicians learn in the field stays with the company.
Bringing the same agent to the technicians who work in the field is under consideration.
Another agent
The first agent built with Omnia reads the transport documents (DDT) of the group’s plants, which arrive by email or from the printer’s scanner. It extracts the data into a table, and doubtful documents go to a queue where a person checks them. It was completed in June 2026 and is in production, and an extension of the platform is under way.
