Models are becoming a commodity. The game is shifting to vertical process knowledge and data governance. And Italy’s mid-market is still largely exposed.
Access to AI is a commodity. The ability to govern it and put it to work on real processes is not
According to analysis published by Bain & Company in 2026, AI budgets keep growing, but returns are slow to arrive: three years after the launch of ChatGPT, very few companies are actually scaling AI beyond experimentation, and 41% point to data integration as the main obstacle. Boards are well aware that 2026 is the year AI must translate into concrete results on the bottom line. AlixPartners, tracking the European software market, arrives at a similar estimate: between 15 and 20% of companies on the continent are equipped to handle the transition. The Italian AI market reached 1.8 billion euros in 2025, with 50% year-on-year growth. This is real growth. But growth in spending doesn’t equate to growth in value generated, and the gap between the two metrics is exactly the problem that anyone operating in this market needs to know how to read.
The emerging consensus among leading industry players is clear: base AI technology is becoming a commodity. Language models are scalable, accessible, and increasingly capable. What’s needed is investment in agentic AI to build autonomous systems at global scale and develop independent architectures. Impresoft, in fact, chooses to design enterprise governance frameworks. A strategy consistent with the size and positioning of the Group and of the client companies it accompanies through digital transformation. The crucial point is clear: the competition is no longer played on the model you choose. It’s played on what you manage to teach it, and on how you govern that knowledge over time.
If technology is a commodity, what isn’t?
The answer lies in process knowledge: not the abstract kind, but the kind layered through years of vertical implementations — in ERPs that know credit approval cycles, in management systems that know how a manufacturing planning chain works or how a clinical diagnostic pathway unfolds. This is the material an AI system has to work with. And it’s material that isn’t generated by buying a model or subscribing to a platform.
At Impresoft, 70% of revenue comes from mid-sized companies. As Group CEO of Impresoft, I get to observe a contradiction that keeps repeating: high interest in AI, almost no governance. Artificial intelligence comes in through the back door: employees use consumer tools, unknowingly sharing sensitive data, internal processes, and strategic information with ungoverned systems. It isn’t a matter of intent. It’s a matter of structure. And companies that don’t police this perimeter are, in effect, handing their information assets over to outside parties.
Data governance: from regulatory constraint to strategic asset
There’s a persistent misconception about data governance: that it’s a compliance matter, a response to GDPR or the AI Act. It isn’t — or not only. It’s the condition that turns AI from a generic service into a proprietary asset. When a company builds a semantic layer over its own data — defining what each term means in its own context, who accesses what, how knowledge gets updated — it’s building something a competitor can’t replicate by buying the same model on the market.
Data without governance is noise. Governed data is the difference between a system that responds and one that understands. And in an economy where AI amplifies everything it finds, the quality of what it finds is the real competitive lever. Whoever doesn’t invest in this layer today isn’t postponing a technology investment: they’re eroding an already-built strategic asset, leaving it unguarded.
Amaltia: the vertical suite as an answer to real problems
Our answer to this scenario is called Amaltia. Not because the market lacked AI solutions — it’s saturated with them — but because it lacked solutions capable of speaking the specific business language of the people who use them, with AI governance by design: granular access policies, automatic versioning, and a full audit trail ensure that every interaction with company knowledge is tracked and verifiable, in full compliance with the European AI Act, without ever ceding control of data to the outside. A CFO doesn’t need a generic AI agent: they need a system that understands the semantics of their margins, the logic of their closing cycle, the anomalies that matter in their chart of accounts. A planning manager at a manufacturing company needs a system that understands the constraints of their supply chain, not one trained on everything and precise about nothing.
This is the difference between horizontal AI and vertical AI. The first lowers the cost of accessing technology. The second generates measurable value on specific problems. Amaltia was born from the premise that companies in manufacturing, retail, fashion, healthcare, and financial services don’t need more technology. They need technology that knows their business and that is able to verify its outputs with the same depth. Impresoft integrates AI into concrete applications and real projects, tailored by sector and need. Amaltia is already in use in the luxury sector, where it has brought AI to the heart of business analysis processes, turning technical documentation into a queryable, always-available intelligence, as well as in automotive, reinventing the digital purchase experience by focusing on conversational commerce and finding a new way of engaging with customers.
And again, in sports, a concrete case is Lega Basket Serie A, which chose Impresoft to develop LUCA, an AI Agent that turns decades of league statistical data into insights, content, and storytelling. Not automation of editorial work: the valorization of an information asset that already existed, but that no editorial team would ever have had the resources to explore systematically.
For Impresoft, the use of AI is concrete and practical. In Magnifica Humanitas, the first encyclical by Leo XIV devoted entirely to artificial intelligence, the Pope writes that technology “is not neutral, because it takes on the face of whoever conceives it, finances it, regulates it, uses it.” It’s a formulation that precisely describes a reality that anyone who runs a company knows well: AI systems are not passive tools. They reflect the choices of those who build them, the priorities of those who adopt them, and above all, the quality of the knowledge they are fed. Governing AI, in this sense, is not just a technology decision: it is a responsibility for the kind of organization you want to build.
Three asymmetries to build, now
Competitive advantage in AI is built on three asymmetries. The first is process knowledge: it isn’t bought on the market, it accumulates over time, in systems, in vertical implementations. Whoever holds it already starts ahead; whoever fails to govern it is giving it away without realizing it. The second is data governance: it isn’t compliance, it’s the condition that allows AI to produce proprietary value — non-fungible, hard to replicate by anyone who doesn’t know the context. The third is verticalization: horizontal AI has a role, but business problems are vertical, and specialization is the only answer that scales without losing precision.
Whoever invests primarily in technology and neglects these three asymmetries is building, at best, a fleeting advantage. At worst, a structural dependency on whoever manages that knowledge on their behalf. The time to build these foundations is now, before the gap between those who have governed their own data and those who haven’t becomes unbridgeable.
Article by Alessandro Geraldi, Group CEO of Impresoft, published on AI4Business: https://www.ai4business.it/intelligenza-artificiale/perche-lai-genera-valore-solo-se-governa-i-dati-aziendali/
