AI in hospitals is not a gadget or an IT project, according to the tenor of a panel discussion at the Health Tech Global Summit in Basel. If you really want to become AI-ready in Switzerland, you have to work on three fronts simultaneously: standardized processes and clean data, financing that honestly prices in the transformation costs, and implementation that does not break down into isolated solutions.
Right at the beginning, one point became clear that is often overlooked in the current AI discourse: the entry into artificial intelligence in the healthcare sector is not a technology decision, but a strategic one. The CEO of Schulthess Klinik, Rodolphe Eurin, argued that AI should not be viewed as a data or IT project. Instead, the key question should be: What results should be improved and what data is needed to reliably measure and optimize them?
In specialized hospitals such as an orthopaedic clinic, the path is somewhat more straightforward: standardized treatment paths are easier to define and results can be collected more consistently. But even there, AI remains primarily a cultural issue. Recording results and continuous improvement are by no means routine everywhere. Without the willingness to make results transparent and learn from them, AI remains an additional layer of complexity rather than a lever for better care.
Operational reality: silos, data chaos and a lack of governance
Matthias Hermann is Head of Health Fund Tenity Group AG, brings experience from Zurich University Hospital and now works in the digital health environment. He painted a sober picture of the current situation, particularly in public hospitals. Many organizations lack standardized, semantically clean data. At the same time, AI expertise in hospitals and IT is often not sufficiently developed to move from pilot projects to operations. This is precisely where many initiatives fail: Models that are convincing in the laboratory do not deliver the same quality in everyday life because data is unstructured in practice and curated data sets are lacking.
Even more serious: there is often no robust clinical and operational data governance. An AI strategy, clear responsibilities, data release processes, monitoring and accountability are homework that needs to be completed before the first rollout. In many hospitals, however, these issues are the responsibility of innovation departments and not the board. As a result, AI remains a collection of experiments rather than a transformation of the operating model towards data and AI-supported management.
Pilot Factory instead of top-down: Why the ecosystem is crucial
Nicolas Loeillot, responsible for the Future of Health Grant at CSS, brought in the external perspective of the pioneers, whereby start-ups, academic teams or innovative companies are often too rarely systematically involved in the Swiss healthcare landscape. He argued that we first need to clarify what we are talking about: AI is not a monolithic block. Radiology, prevention, coaching, administrative automation or classic machine learning applications have different levels of maturity, risks and implementation paths.
His approach is based on a bottom-up mechanism: a pilot factory that prepares and structures start-ups and brings them to hospitals not as a PR demo, but as a measurable proof of value. Three axes are decisive here: the economic effect (does the solution save costs in the system?), the clinical-operative effect (is it actually accepted by specialists and patients?) and the scalability (is the success reproducible or just an isolated case?). It is precisely this measurement logic that should make it easier for hospital management to finance innovation not out of principle, but on the basis of comprehensible results.
Financing dilemma: the price of transformation is underestimated
At the latest when it came to the topic of financing, the tension that Swiss hospitals are currently facing became clear. Prof. Dr. med. Katrin Hoffmann, CEO Hoffmann Global Health Advisory, referred to the pressure in reimbursement-driven everyday life: tariffs are under pressure, while cantonal policy sometimes pushes investments in buildings more strongly than investments in digital infrastructure. There is also a classic governance conflict: those who understand technology do not necessarily decide on budgets and those who control budgets are often skeptical about new technologies.
Her reference to hidden costs was central. If you want to introduce AI seriously, you have to reckon with more than just license costs: Employee training, workflow redesign, data cleansing, model performance monitoring, hardware and infrastructure requirements, IT capabilities and value creation mechanisms are all part of the business case. However, many boards of directors calculate costs per feature instead of costs per result. It is precisely this change in perspective that determines whether AI is seen as a permanent construction site or a strategic investment.
Private logic: efficiency, yield and competition over quality
Eurin put it pragmatically: every innovation must either reduce costs or increase revenue. AI can increase efficiency potential in the administrative area, but also in the medical core. For him, competition on results is particularly important: those who demonstrably achieve better results gain market share and can thus refinance investments.
Things got exciting when the discussion went beyond the individual hospital. Outcome-based remuneration models could create a mechanism that rewards quality improvement. Particularly in areas where higher quality tends to reduce costs (fewer complications, fewer revisions, fewer readmissions), closer integration between service providers and payers would be a possible way forward, provided that the results are consistently collected and accepted as a control parameter.
Creative sources of capital: When tech providers become research partners
Herrmann introduced a perspective that is still too little visible in the Swiss debate: large tech providers have a vested interest in scaling AI in hospitals, for example through increasing cloud or data center consumption. This could give rise to new co-financing models in which technology providers (co-)finance pilot projects because they benefit from higher usage in the long term. He drew a comparison with research, where industry partners have been providing infrastructure, instruments or technology capacities for years to make studies possible in the first place.
This is not a sure-fire success: such models require transparency, clear rules on conflicts of interest and governance that protects clinical priorities. But as an approach to lower investment barriers, this “out-of-the-box” financing could increase in the coming years.
Where AI works first: Back office, control and prevention as frontier
Three priorities emerged in the specific fields of application. Firstly, administrative and operational processes in the back office. AI is seen as an easy target to achieve here because regulatory hurdles are lower and business cases are simpler. Topics such as surgery planning, resource management, MR utilization or automated reconciliations in billing and administrative processes often tie up highly qualified working time in Excel lists and manual click paths. AI agents could quickly achieve measurable effects here.
Secondly: quality, patient safety and cognitive relief. Prof. Hoffmann emphasized that AI should not mean another screen and even more clicks. Its value lies in reducing the cognitive load in busy teams, avoiding redundancies, improving throughput times and increasing the predictability of hospital operations, always linked to clinical goals.
Thirdly, prevention as a strategic zone beyond the hospital. Loeillot formulated the most ambitious vision from an insurer’s perspective here: the monetization of prevention and a new promise to insured persons, namely not to become a patient in the first place if possible. Preventive AI, coaching, lifestyle data and early interventions could be a field that today lies between mandates and is therefore considered a frontier.
The elephant in the room: system data and Swiss slowness
One point of discussion remained present as a background noise: the systemic data situation. If patient data is still partly paper-based and held by GPs, if national electronic patient dossiers are making slow progress and data protection debates are slowing down scaling, then the AI horizon is shifting. The comparison with countries such as Denmark, which have nationwide EHR structures (electronic health records), showed the consequence: those who can use data at population level will translate AI into real health gains more quickly.
At the same time, the technical reality was also mentioned: The state of data in hospitals, insurance companies and care structures is often a real mess. In data science projects, the majority of time is spent on data cleansing and preparation. Without this basis, AI-ready remains a buzzword.
AI-ready is a conversion, not an addition
The panel discussion made it clear: throughout Switzerland, readiness for AI is not determined by individual tools, but by infrastructure in the broadest sense, such as process standardization, data governance, results orientation, financing including transformation costs and an implementation logic that prevents isolated solutions. If you really want to use AI, you have to anchor it in the operating model and take responsibility for it at board level.
And perhaps that is the most important message from Basel: the volume of the debate is of little help. The quiet, often invisible work on the foundations is crucial.
Binci Heeb
Read also: Amal Clooney: AI as a lever for access to justice
Search:
Sponsors:
