Artificial intelligence is entering the Italian healthcare system. It is doing so along at least two complementary trajectories: on the one hand, a national platform promoted by the National Agency for Regional Health Services (AGENAS) to support general practitioners (GPs); on the other, a significant number of healthcare organizations that are combining the purchase of market technologies with the in-house development of AI solutions.
Two chapters of the Rapporto OASI 2025 describe a system in ferment. One chapter, written by Giulia Broccolo, Francesca Guerra, Giulio Guidotti and Francesco Longo, analyzes the attempt to introduce an AI platform to support primary care on a national scale, one of the first of its kind in Europe. The other, written by Vittoria Ardito, Giulia Cappellaro, Amelia Compagni, Francesco Petracca and Luigi Maria Preti, enters the “laboratories” of four major Italian hospitals to understand why and how they decided to become developers of artificial intelligence.
The two analyses suggest that innovation works when it starts from real clinical needs, is equipped with adequate skills, and chooses an organizational dimension that cannot be limited to the IT department. Developing an algorithm in a laboratory is relatively simple. Bringing it into everyday clinical practice, integrating it, certifying it, and ensuring its daily use is a leap in complexity that requires a multi-level development and change management strategy.
Three levels of response
The first study focuses on the AI platform promoted by AGENAS within the framework of the National Recovery and Resilience Plan (PNRR), with the aim of providing GPs with a clinical decision support system capable of guiding diagnosis, prevention, and the management of chronic conditions in high-prevalence areas (diabetes, pulmonology, cardiology, oncology and neurology).
Before this initiative, AI adoption within the National Health Service (SSN) was fragmented and often limited to individual departments or organizations. A systemic and coordinated initiative at the national level was lacking.
The study’s questions were:
- Micro level (professionals): how do general practitioners perceive AI? Do they see it as support or as a threat to their autonomy?
- Meso level (organizations): how do healthcare managers intend to govern implementation? With which change management levers?
- Macro level (policy): which regulatory and governance choices can foster adoption?
The chapter on healthcare organizations, by contrast, asks which factors lead organizations to develop AI solutions internally and which, instead, drive them to purchase them from the market. The state of the art shows a strong prevalence of off-the-shelf solutions, especially in diagnostics (for example, in radiology). However, some organizations, and in particular the Scientific Institutes for Research, Hospitalization and Healthcare (IRCCS), have begun to build internal data science and AI capabilities, creating dedicated centers.
Achieving critical mass
In the case of the AGENAS platform, 21 GPs considered key opinion leaders were interviewed. Their attitude is not ideologically hostile. Many see AI as a way to lighten the bureaucratic burden and recover clinical time, but several recurring concerns nonetheless emerge:
- The cognitive atrophy of clinical reasoning,
- Algorithmic opacity.
- Medical-legal liability.
- Privacy.
- The need for integration into existing information systems, to avoid having to manage yet another portal.
For their part, healthcare managers believe that the benefits of the AGENAS platform primarily concern appropriateness of care, followed by proactive management of chronic patients, and only marginally cost containment.
In hospitals that develop AI internally, models almost always originate from clinical needs expressed by professionals themselves. AI should not be imposed; it must emerge as a response to unmet clinical needs, often related to complex or rare conditions.
The research adopted a multiple case study design across four institutions: Humanitas Research Hospital; Fondazione Policlinico Gemelli; San Raffaele Hospital; and Policlinico Sant’Orsola in Bologna, which are adopting different organizational solutions.
Humanitas has created a dedicated AI Center with around 50 professionals; Gemelli separates industrial valorization activities into an ad hoc company (GDMH); San Raffaele is building a proprietary platform in partnership with Microsoft; Sant’Orsola integrates data science and IT within a single organizational structure. While the models differ, all experiences tend to concentrate interdisciplinary skills and oversee the entire data lifecycle.
Not all Italian institutions can afford to develop solutions ready for clinical practice; a critical mass of data, skills, and resources is required. In a market where data science and AI skills are scarce and highly sought after, the public sector may face significant difficulties in attracting adequate personnel, and development may stall at the experimental level without the support of external actors. Therefore, in the traditional make-or-buy dilemma, organizations appear to be settling on a “make and buy” approach, in which development is internal when customization is needed or the market does not offer mature solutions; where speed and ready-made certifications are required, the market is used.
The constraint of the labor market
From the intersection of the two studies, several managerial implications emerge.
- Scale matters. Internal development works where there is critical mass, scientific vocation, and interdisciplinary skills. Not all organizations can, nor should, follow this path. At the same time, a national platform can guarantee equity and standardization, but it must be able to adapt to local contexts.
- Skills are the real structural constraint. All hospital cases show significant investment in data science, engineering, and bioinformatics skills. In the public sector, contractual rigidities make it more complex to attract these profiles. Without adequate human capital, AI remains experimentation.
- Openness to other actors is necessary. Co-design, training, feedback loops, and collaborative governance emerge as enabling conditions. AI does not replace the physician, but redefines the role: from isolated decision-maker to mediator between scientific evidence, algorithm, and patient values.
Ultimately, the two studies show that artificial intelligence in healthcare is a test of institutional maturity. Those who govern scale, skills, and professional identity truly govern innovation.
Giulia Broccolo, Francesca Guerra, Giulio Guidotti, Francesco Longo, “Una piattaforma di Intelligenza Artificiale a supporto dell’assistenza primaria: impatti su pratica e identità clinica, modelli organizzativi e agenda di policy.” In Rapporto OASI 2025.
Vittoria Ardito, Giulia Cappellaro, Amelia Compagni, Francesco Petracca, Luigi Maria Preti, “Le scelte di sviluppo di soluzioni di intelligenza artificiale nelle aziende sanitarie italiane.” In Rapporto OASI 2025.
