Skills come before servers

In 2025, 16.4% of Italian companies with at least ten employees used at least one artificial intelligence technology. It was 8.2% in 2024 and 6% in 2023. The figure doubled in a year, but it remains below the European average (19.95%) and is unevenly distributed: 53.1% among large companies, 27% among medium-sized ones, 14.2% among small ones.

What holds companies back

ISTAT asked the companies that had assessed artificial intelligence without adopting it which obstacles they had met. In first place, at 58.6%, is the lack of adequate skills. Then come regulatory uncertainty (47.3%), the availability and quality of data (45.2%), privacy concerns (43.2%) and costs (43.0%).

The order deserves attention. Cost, which is the most immediate objection, is the last of the five obstacles. The first concerns not the technology but the people who are supposed to use it.

The Politecnico di Milano's Osservatorio adds a figure on small and medium-sized enterprises: 76% do not invest in artificial intelligence and do not plan to, and only 7% have structured training on the subject. Among the companies that use it, the most widespread technology is the generative kind (59.1%), that is, the assistants that write, summarise and answer: the tool that more than any other depends on who uses it.

Why training comes first

A language model is not management software. It has no list of functions to be learned once and for all, and the quality of what it produces depends on how it is questioned and on who reads its answers. Three consequences.

Results have to be assessed. An assistant always answers, even when it does not know. Wrong answers are written with the same confidence as right ones. Recognising them requires competence in the subject matter and a habit of verification that cannot be improvised.

Daily use touches confidential data. Whoever pastes a contract into a free service to have it summarised has just transferred that contract to an external provider, on terms they have not read. In most cases it is not a decision: it is a gesture. Knowing what can be shared, and with which tool, is the first security measure, and it comes before any investment in infrastructure.

The value lies in each person's work. The useful applications of artificial intelligence in a company are not found by the supplier: they are found by those who know the work. An accounts clerk reconciling invoices, a technician writing manuals, a salesperson preparing quotations know where they lose time. If they also know what the tool can do, they find the right use on their own.

An obligation, not just advice

Article 4 of the European regulation on artificial intelligence (AI Act) has been in force since 2 February 2025 and also applies to those who use artificial intelligence systems, not only to those who produce them. After the July 2026 revision (the so-called Digital Omnibus, Regulation 2026/1744) the article requires companies to adopt measures to promote AI literacy among their staff, assessed according to the context. It does not impose a guaranteed level for every person, but the obligation remains, and training is the simplest way to document it.

For professionals there is also Italian law 132/2025, in force since 10 October 2025: article 13 requires accountants, lawyers and the other members of a professional order to inform their clients about the artificial intelligence systems they use, in clear and comprehensive language. It is difficult to inform about what one does not know.

What a day on site covers

The training we offer takes place on site, with the people who will use the tools, on their actual work.

• What a language model does and does not do, with examples taken from the company's documents.
• Safe use in daily work: what can be shared, with which tool, and what must stay within the company.
• How to read an answer: checking sources, recognising errors, the limits of the tool.
• What the law says, in half a page: AI Act, Italian law 132/2025, GDPR.
• A list of concrete uses for each department, built with the people who work there.

The server, if needed, comes afterwards. And often, after a day like this, the company already knows whether it needs one.

Sources

• ISTAT, Imprese e ICT, anno 2025, press release.
• Eurostat, Use of artificial intelligence in enterprises, December 2025.
• Osservatorio Artificial Intelligence and Osservatorio Innovazione Digitale nelle PMI, Politecnico di Milano, 2026 results.
• Regulation (EU) 2024/1689 (AI Act), art. 4, as amended by Regulation (EU) 2026/1744.
• Italian law no. 132 of 23 September 2025, art. 13.