Healthcare and Artificial Intelligence in Lombardy: a Half-Baked Revolution
In theory, artificial intelligence is the future of healthcare. But in the corridors of Lombardy's hospitals, the future is advancing at different speeds. This is one of the conclusions of new research by Bocconi University, which has mapped for the first time the adoption of AI applications in clinical practice in public and private facilities in Lombardy.
The study, entitled "Adoption of artificial intelligence applications in clinical practice: Insights from a Survey of Healthcare Organizations in Lombardy, Italy," published in Digital Health and authored by Vittoria Ardito, Giulia Cappellaro, Amelia Compagni, Francesco Petracca, and Luigi M. Preti, paints a multifaceted picture: of the 46 organizations that responded, as many as 57% have not yet adopted any AI applications in the clinical setting. The survey was conducted in 2024. Although AI is showing a rapid adoption curve, the results allow for some general reflections.
Three approaches, three different healthcare systems
The sample's responses on how AI is being adopted in clinical practice reveal three distinct strategic approaches organizations adopt. Only six organizations (13%)—mainly research-oriented IRCCS (Scientific Institutes for Research, Hospitalization, and Care)—are developing their own solutions, often in collaboration with academic or technological partners. A second, larger group (14 organizations, 30%) relies on the purchase of CE-marked commercial applications already available on the market. The third and largest group consists of 26 organizations (57%) that have not yet adopted any solutions. The trade-off between purchasing and developing innovative solutions is also emerging in healthcare organizations.
"Most of the applications in use today are still linked to diagnostic imaging, where the commercial offering is broad and well-established," explains Giulia Cappellaro, associate professor of Public Management at Bocconi University. "The most innovative applications, for example for prognosis in chronic patients, are often still in the development phase."
Radiology reigns supreme, chronic conditions lag behind
A total of 56 applications in use or in the testing phase were surveyed. Fifty-four percent support diagnosis, 48% support prognosis, and only 11% focus on treatment optimization. Radiology is the clinical area with the highest concentration (30% of applications), followed by oncology and diabetology.
"We are still a long way from widespread and structured adoption," warns Amelia Compagni, director of Cergas Bocconi. "In many cases, the implementation of AI is driven by the initiative of individual clinicians or departments, without a real strategic plan or dedicated structures. This poses a serious limitation to scalability."
Governance, training, and cultural barriers
Only the group of internal developers has implemented formal roles and structures to govern AI adoption (67%). The others rely on informal or ad hoc initiatives. Training? There is some in only 33% of cases in the first group and even more sporadic elsewhere.
When it comes to obstacles, perceptions vary depending on the organizational profile. Development organizations cite privacy, interoperability, and lack of economic incentives as the main barriers. Those purchasing commercial technologies mainly mention organizational culture and a lack of trust among operators. Non-adopters report a shortage of qualified personnel as the main obstacle.
"We need a system strategy to guide organizations through this momentous transition," emphasizes Vittoria Ardito, lecturer at SDA Bocconi. "AI can truly improve care pathways, but without investment in governance, skills, and digital infrastructure, it may stay the preserve of few centers of excellence."
An agenda for the future: what is really needed to get AI off the ground in healthcare
The Bocconi study does not merely provide a snapshot of the current situation. It also offers a compass for those responsible for driving change, from regional policymakers to healthcare managers.
The first step is to break out of the silo mentality. "We need significant investment in collaboration between healthcare facilities, universities, technology providers, and institutions," emphasizes Amelia Compagni. "Joint projects, regional platforms for data sharing, stable alliances between IRCCS and local healthcare companies: without a coherent ecosystem, innovation risks remaining confined to few centers."
The second pillar is training. Digital skills are still too unevenly distributed. Structured programs are needed for clinicians, managers, and engineers, not only on what AI does, but on how to truly integrate it into care processes. Specific training on individual applications has proven decisive in the success stories mapped.
Finally, the authors propose developing clear evaluation and governance frameworks that go beyond clinical effectiveness and also consider organizational impacts, economic sustainability, and acceptability by operators and patients. "The adoption of AI cannot depend on haphazard initiatives or commercial pressure," warns Vittoria Ardito. "A shared strategy is needed, with clear criteria for prioritization and funding."
In summary: more collaboration, more skills, more rules. Only then can artificial intelligence truly become a structural resource for healthcare in Lombardy—and beyond.
Vittoria Ardito, Giulia Cappellaro, Amelia Compagni, Francesco Petracca, Luigi M Preti, "Adoption of artificial intelligence applications in clinical practice: Insights from a Survey of Healthcare Organizations in Lombardy, Italy", Digital Health, Vol. 11, DOI https://doi.org/10.1177/20552076251355680
