AI adoption: where use ends and integration begins

Walk into a large Italian company and ask who uses artificial intelligence, and plenty of hands go up: marketing for copy, IT for code, someone in finance for analysis. Ask instead who decides which of those uses enter official processes, on what data and under whose responsibility, and the room empties. That gap, between who uses AI and who governs it, is where the whole question of adoption sits.

AI adoption is often confused with its diffusion, and the two are far apart. Diffusion is measured in licences activated and people who open a model every morning. Adoption is measured in how much that model has changed the way a company produces value and answers for what it produces.

The difference between using AI and adopting it

An organisation has adopted AI when it can do three things: measure what AI produces, decide where that contribution enters its processes, and answer for the decisions that follow. Until those three are in place, AI stays a tool in the hands of individuals and never becomes an organisational capability. A company where two hundred people use a generative assistant every day, yet nobody can say what changed in the accounts or who approved which uses, has diffusion at its peak and adoption at zero. It is the most common situation, and also the most expensive, because it looks like progress without being it.

The four capabilities that hold adoption up

Adopting AI is an organisational capability that rests on four foundations, built together. Readiness comes first: usable data, real skills, processes able to absorb AI, a culture that does not push it away. Delivery comes next, the ability to take a project from pilot to production and to scale it, and this is where most initiatives die. On risk sits governance, with a clear line of accountability over who decides and who supervises, and alignment with Italy's Law 132/2025 and the EU AI Act. Confidence remains, the commitment of leadership and the trust of the people who do the work, without which even a system that works goes unused. These are the four dimensions the AI Rating measures, and finding all four mature in the same company is rare.

71% have a project, 9% have governance

The Italian numbers describe exactly this imbalance. In 2025 the Italian AI market reached 1.8 billion euro, up 50% on the year before, and 71% of large companies had launched at least one AI project, rising from 59% in 2024, according to the Artificial Intelligence Observatory of the Politecnico di Milano, which presented its research on 5 February 2026. 84% use ready-to-use Generative AI tools. Beneath those figures the picture shifts: only one large company in five uses AI pervasively across several functions, and just 9% report fully integrated AI governance. Among small and medium enterprises the share that has launched projects drops to 8%. The same Observatory notes that eight workers in ten use AI tools their employer does not provide, so much of real adoption happens outside any perimeter and any decision. Italy has spread AI wide and integrated it deep in only a minority of cases.

The cost of staying on the surface

Stopping at diffusion has a price, paid on several fronts. The first is economic: money spent on access that never enters processes stays an individual productivity gain and never becomes an advantage for the organisation. Then there is defensibility, because individual use is quick to copy while integration into processes is not, and the distance between those who have governed adoption and those who have not widens over time. On the regulatory front, with Law 132/2025 in force and the AI Act coming into effect, ungoverned adoption becomes exposure, because it ties the company to decisions taken by models it has no record of. Above all sits a dependency of sequence: automation and agents are built on governed adoption, and a company that has not been through this stage has no foundation for the ones that follow.

Measure maturity before scaling it

It starts with knowing where you actually stand. Before choosing where to invest, a company needs an objective measure of its capacity to adopt AI across the four dimensions, from readiness to confidence. This is the logic of the AI Rating, the assessment with which ZeroFive.AI captures a company's AI maturity on objective criteria and returns the gaps before they turn into costs. From that measure the concrete choices follow: where it makes sense to invest, and what to validate before putting it into production. Measure, decide, validate, in that order.

The 9% that hold integrated governance today start from the same technology as everyone else; they decided earlier, and on measurable ground. The question to bring to the board is not how much we are spending on AI, but how much of that spend we can actually govern. If the answer is not clear, the first step is to measure it.

It starts with a conversation: hello@zerofive.ai