In 2020, Italy's artificial intelligence market was worth 300 million euro. By 2025, it had reached 1.8 billion. Six times in five years, with growth rates no other technology sector recorded in the same period. Yet, looking at Italian SMEs, which make up 90% of the country's productive fabric, the picture is more complex: only 8% have launched a structured AI project, compared to 71% of large enterprises.

The question every business leader should be asking is no longer "should we adopt AI?". It is: are we adopting it in the right way? And the right way has two necessary conditions: governance and measurable value.

Investment is there. Value extraction, not yet.

The Italian market figures tell a story of real acceleration. From 2022 onward, annual growth never fell below 32%, reaching 52% in 2023 and 58% in 2024. Generative AI lowered the barrier to entry, made tools accessible and triggered a first wave of broad experimentation across companies of all sizes.

But most of this spend concentrated where adoption is easiest: marketing and content, document automation, customer service. Useful areas, but peripheral to the decisions that drive EBITDA. Pricing, portfolio mix, raw material procurement, operational planning: these remain managed with the same tools as ten years ago, despite being the areas where AI ROI is systematically highest.

The problem is not technological. It is methodological.

Three recurring errors explain why 72% of companies that adopt AI fail to extract measurable value. The first: AI is deployed in the wrong place, far from the decisions that move the margin. The second: tools are not connected to real company data. They do not see the ERP, do not read the order history, do not know actual costs. The result is recommendations that are plausible in form and useless in substance.

The third error is the most subtle: opacity. 67% of managers who abandon AI tools cite as the primary reason not errors in the recommendations, but the inability to explain those recommendations to leadership. A system that is correct but incomprehensible has no operational value for someone who must justify a decision to the board.

The real challenge: from adoption to governance.

The market is growing, the technology is available, costs are falling. The gap that separates value-creating SMEs from those accumulating costs is not access to tools: it is the capacity to integrate AI into defined processes, with clear ownership and KPIs measured from day one.

This means designating an "AI decision owner" for each critical area, not leaving adoption to the individual initiative of separate departments. It means connecting tools to real company data before expecting useful recommendations. It means defining 3 to 5 impact indicators before go-live, not after. Companies that measure from the start are 2.4 times more likely to scale the project beyond the pilot phase.

Demonstrating ROI: the non-negotiable condition.

The average ROI of AI projects in European SMEs is 340% over three years, with breakeven in 8 to 14 months. But this average conceals a highly asymmetric distribution: those with a method generate results; those running disconnected experiments do not.

The areas with the highest return, document automation (500% over 3 years), structured pricing (420%), predictive sales analytics (380%), are often the least addressed. Not because the ROI is not demonstrable, but because it requires integration with real data and systematic measurement that most SMEs have not yet built.

The challenge for the next 12 to 24 months is not persuading Italian SMEs to invest in AI. Investment is already growing. The challenge is turning that spend into measurable competitive advantage: AI embedded in processes with defined ownership, connected to the data that matters, evaluated against real KPIs. Those who build this architecture today accumulate an advantage that compounds over time. Those who wait accumulate a gap that becomes progressively more expensive to close.