In recent years, Artificial Intelligence has become a fixture in the language of business. First as a topic for experimentation, then as a lever for efficiency, and today increasingly as a potential driver of transformation for processes, skills, and organizational models.

The real question, though, is understanding what “adopting AI” actually means. Using a single generative tool, introducing automation in a few departments, or rethinking business processes around data, governance, and skills — these represent very different levels of maturity.

This is where the debate on enterprise innovation gets more interesting. It’s no longer enough to ask how many companies are using AI. The more useful question is: how many are managing to turn it into operational, measurable, and sustainable value?

The research reviewed points to a consistent picture: AI is spreading fast across enterprises, but adoption is outpacing organizations’ ability to govern it. The data show clear growth in Europe and in Italy, alongside a still-significant gap between tool usage, digital maturity, internal skills, and process integration. The real challenge, then, isn’t just introducing AI — it’s making it part of a more mature, measurable, and sustainable operating model.

AI adoption in enterprises: what the European and Italian data show

At the European level, AI adoption among enterprises is growing. According to Eurostat, in 2025 19.95% of EU enterprises with 10 or more employees use AI technologies. Size plays a big role here too: AI is used by 17% of small enterprises, 30.36% of medium-sized enterprises, and 55.03% of large enterprises. Eurostat notes that this gap can be explained by factors such as implementation complexity, economies of scale, and investment costs.

The Italian picture fits this same trajectory. The ISTAT report shows that in 2025, 16.4% of Italian enterprises with 10 or more employees use at least one AI technology, up from 8.2% in 2024 and 5.0% in 2023. Growth is even more pronounced among large enterprises, which went from 32.5% in 2024 to 53.1% in 2025; among SMEs, usage doubled, from 7.7% to 15.7%.

These figures confirm a clear trend: AI is no longer confined to a handful of experimental cases or large tech groups. At the same time, they reveal a still-significant gap between large and smaller enterprises. It’s not just a matter of access to tools, but of the ability to integrate them into organizational contexts that often differ widely in resources, skills, and digital maturity.

Overall digitalization levels also help make sense of this. Still according to the same ISTAT survey, in 2025 nearly 80% of Italian enterprises with 10 or more employees reach a basic level of digitalization, defined as adopting at least four out of twelve digital activities under the Digital Intensity Index. 38.1% reach at least a high level, meaning at least seven out of twelve digital activities. Among large enterprises, these figures rise to 96.4% and 81.4% respectively.

The ITIR – University of Pavia research adds another layer, focusing on medium-to-large Italian enterprises. In the sample analyzed, 59.8% of the workers surveyed report making at least some use of AI in their work. This varies sharply by role: adoption is highest among Top Managers (91.2%), among those who describe themselves as AI experts (89.7%), and among younger workers (71.9%).

Three things stand out:

  1. AI is growing rapidly, including in Italy;
  2. large enterprises remain ahead of SMEs;
  3. basic digitalization is widespread, but more advanced maturity remains less common.

It’s in this gap between adoption and maturity that the more interesting discussion takes shape.

Why adopting AI doesn’t yet mean transforming the company

Internationally, several studies converge on one point: using AI doesn’t necessarily mean transforming the organization. The McKinsey report finds that companies are starting to build structures and processes to generate value from GenAI, but the journey is still in its early stages. Practices cited include redesigning workflows, strengthening governance, and bringing in senior roles to oversee AI.

BCG also stresses the gap between potential and impact. According to their research, only 5% of the companies analyzed are classified as future-built — capable of generating AI value at scale; 35% are scaling AI and starting to generate value; the remaining 60% see only limited material benefits, despite their investments.

Italian evidence helps make this gap more concrete. Research from the Politecnico di Milano’s AI4Innovation Observatory distinguishes between occasional and structured use of AI in innovation processes. Occasional use is experimental, individual, and not integrated into processes; structured use involves defined workflows, dedicated tools, governance, and metrics. It’s precisely between these two modes that the real transformation of the innovation process plays out.

The ITIR – University of Pavia study confirms the same gap from a different angle. In the medium-to-large enterprises analyzed, AI is already widespread, but only 14.4% of workers feel they have meaningful competence in using these technologies. Moreover, just 1.8% of companies sit at a very advanced, deeply embedded stage of AI maturity. The data show that, despite growing diffusion, the process of organizational adoption often remains immature.

This distinction is central. One company may use generative tools to write text, analyze documents, or support day-to-day tasks without having genuinely changed how it works. Another may embed AI into core processes, defining responsibilities, metrics, controls, and adoption pathways. In both cases, you could say AI is “in use” — but the level of transformation is very different. The difference, then, isn’t the presence of the technology, but the organization’s ability to make it a stable part of how it operates.

AI and innovation processes: where usage is most mature

International research shows that the value of AI depends less and less on simple access to tools, and more and more on the ability to embed them into processes. McKinsey, for instance, links value generation to the capacity for rewiring — rethinking how the organization works: workflows, governance, adoption, training, KPIs, and trust in outputs.

In the Italian context, the AI4Innovation Observatory report is particularly useful because it looks at how GenAI and Agentic AI enter the different stages of the innovation process, distinguishing between occasional and structured use.

In the front-end — the activities that precede a project’s formalization — AI is most widespread in the stages where experimentation is easiest. Idea generation is the area with the highest overall diffusion: 45% of companies report occasional use and 13% structured use. PoC building also shows a relatively more advanced level, with 18% structured use. By contrast, idea evaluation remains the least structurally supported stage, at 9%.

The pattern is clear: AI takes hold more easily where operational barriers are low, such as idea generation or prototype building. It struggles to become structural in stages that require greater trust in outputs, stronger governance, and review of decision-making criteria.

In the back-end of innovation — the stage where selected ideas are turned into concrete solutions — the picture is similar. Knowledge management is the most advanced area: about 60% of the sample reports at least some form of AI use, with 27% structured and 33% occasional. In decision making, on the other hand, fewer than one company in three has started integrating AI, and structured use sits at just 13%. In project management, 70% of the sample uses no AI tools to support project management, and only 12% report structured use.

The ITIR – University of Pavia report helps explain this gap from another angle: AI often enters organizations from the bottom up. 6.5% of respondents say they use paid AI applications at work funded out of their own pocket. This figure is higher among Experts, at 11.5%, and among profiles defined as Rebels, at 9.6%. It’s a sign of spontaneous, not always governed, adoption — one that can anticipate real needs, but also raises questions around security, data governance, and process standardization.

In short, AI is most easily used to search for information, generate ideas, summarize documents, or build prototypes. It becomes harder to integrate at the moments when an organization needs to make decisions, monitor projects, assign responsibilities, and change established operational routines.

Skills, governance, and data: the conditions for using AI well

Internationally, the skills issue is especially visible among SMEs. The OECD report describes AI adoption among SMEs as still relatively low compared to other digital technologies and to large enterprises. The paper identifies connectivity, data, algorithms, computing capacity, skills, and finance as key enabling factors.

In Italy, the problem shows up very concretely. According to ISTAT, enterprises that don’t use AI but have considered adopting it point to five main obstacles: a lack of skills (58.6%), insufficient regulatory clarity (47.3%), unavailable or poor-quality data (45.2%), privacy and data protection concerns (43.2%), and high costs (43.0%).

The AI4Innovation Observatory report also confirms how central skills are. 96% of respondents believe new skills need to be developed within innovation teams to work effectively with GenAI and agents — 52% significantly so, 44% at least in part. To close these gaps, companies rely mainly on internal training (75% of respondents), followed by on-the-job upskilling through pilot projects (62%) and partnerships with vendors and consultants (48%).

The data on sought-after profiles is equally interesting: 51% of respondents prioritize hybrid profiles — people who combine innovation management skills with a baseline understanding of AI. Purely technical profiles, such as data scientists, ML engineers, or AI engineers, come in at 32%.

The ITIR – University of Pavia research adds a complementary data point on governance. 50.5% of enterprises run training activities, workshops, or events dedicated to AI. More structured solutions are far less common, though. Only 17.0% of the sample has developed rules, regulations, or processes specifically designed for working with AI. Just 16.3% has created dedicated AI roles, and a mere 8.6% has an organizational unit fully dedicated to it. Formal governance and usage policies for AI are in place at 21.1% of organizations.

In short, companies don’t just need new tools. They need to build the conditions to use them well:

  • widespread skills, not just specialist ones;
  • clear governance over data, risk, and responsibility;
  • processes redesigned around genuinely useful use cases;
  • metrics that distinguish experimentation, expected productivity, and actual value generated.

This is a meaningful distinction. AI can simplify repetitive tasks, speed up information analysis, and support operational decisions. Introduced without governance, though, it can have the opposite effect: disconnected tools, parallel workflows, duplicated data, untracked usage, and decisions that are hard to trace back.

AI investment: growth, expectations, and process integration

Internationally, AI is now firmly established among investment priorities. The Deloitte report notes that AI is moving from a phase of experimentation to one of deeper enterprise integration. Globally, 25% of respondents say AI is having a transformative effect on their company, 30% of organizations are redesigning key processes around AI, while 37% say they’re still using it at a surface level.

In the Italian market, Deloitte finds that 82% of the companies surveyed plan to increase their AI investments over the next year, and 92% expect a productivity increase from adopting these tools. These are useful figures, but should be treated with caution: they measure the expectations, perceptions, and investment intentions of surveyed companies, not definitive proof of economic impact already realized across the board.

The AI4Innovation Observatory research adds a useful detail on the nature of investment among the Italian companies analyzed. The most frequently cited item is employee access to LLMs and generative tools, cited by 73% of respondents. Next is the in-house development of custom solutions or solutions based on proprietary data, cited by 67%. Vertical tools with built-in AI are cited by 38%, low-code and no-code platforms by 29%, while only 9% report not investing in AI at all.

The ITIR – University of Pavia report offers a more cautious figure on actual, consolidated investment. Among the medium-to-large Italian enterprises in the sample, 17.8% report making some form of AI investment, equal to at least 1% of their total budget. The report also notes that nearly 55% of the sample either doesn’t know, or prefers not to disclose, the share of budget allocated to AI. This may reflect confidentiality concerns, but also limited spend traceability — consistent with AI adoption that is often hybrid and bottom-up.

This distribution points to two different levels of adoption. On one hand, access to general-purpose tools is often the first step — useful for spreading familiarity and boosting individual efficiency. On the other, developing custom solutions or ones based on proprietary data marks a more strategic shift: value doesn’t come only from access to the model, but from the ability to connect it to the organization’s own data, processes, and knowledge.

The stated priorities confirm this framing. According to the AI4Innovation Observatory report, 72% of respondents name increased operational efficiency as a priority, and 59% name automating repetitive tasks. 43% cite exploring new opportunities and new business models, while 40% point to improved decision quality thanks to a broader information base.

The point, then, isn’t just how much is invested, but where it’s invested and how deeply. Access to tools is necessary, but not enough to build a lasting advantage. The real difference lies in the ability to connect AI to proprietary data, core processes, internal skills, and governance.

From adoption to governance: the organizational challenge of AI

The sources reviewed tell a fairly clear story. AI in enterprises is growing fast, including in Italy. Large companies are ahead, SMEs are accelerating but remain more exposed to constraints in skills, resources, and digital maturity. Investment is rising, expectations are high, but real transformation still depends on the ability to integrate AI into processes, data, governance, and skills.

The AI4Innovation Observatory research reinforces this reading, since it shows, at the level of individual innovation processes, the gap between occasional and structured use. AI is already present in many activities, but often as ad hoc support, mostly through general-purpose tools. Moving toward deeper integration requires a roadmap, skills, metrics, workflow redesign, and governance capable of guiding adoption over time.

The ITIR – University of Pavia report adds a decisive point: new forms of AI tend to enter companies even without explicit direction, through spontaneous initiatives, local experiments, and bottom-up practices. This is why the question isn’t just whether to adopt AI, but how to govern its spread before it becomes opaque, fragmented, or inconsistent with the organization’s goals.

In the coming years, the real divide probably won’t be between companies that “use” AI and those that don’t. It will be between organizations that introduce it as an add-on tool, and organizations that manage to embed it coherently into their operating model.

It’s a less visible distinction, but a far more concrete one. Because innovation, in business, isn’t measured only by the technology adopted. It’s measured by the ability to make it useful, governable, and sustainable over time.

Bibliography

A note on sources

The sources cited don’t all measure the same phenomenon. ISTAT and Eurostat offer a statistical snapshot of AI adoption among enterprises. Politecnico di Milano and the University of Pavia take a closer look at maturity, innovation processes, and governance. McKinsey, BCG, Deloitte, and the OECD provide a managerial and international reading of the phenomenon. For this reason, the figures shouldn’t be read as directly comparable percentages, but as different vantage points on the same transformation.