# Why Clinical AI Should Not Be Procured Too Early
Startups, Hospitals — and Switzerland vs. the Netherlands under EU Sovereignty
Hospitals across Europe face the same fundamental question: how should clinical AI be introduced responsibly?
The default answer is often: run a pilot, then procure a solution.
For clinical AI — especially systems that reason across complex patient context — this approach is premature.
The contrast between Switzerland and the Netherlands, and their different relationship to EU sovereignty, makes this particularly visible.
---
Clinical AI Is Learned Into the Clinic
Clinical AI is not a static product.
Its behavior depends on:
- —local workflows
- —institutional norms
- —patient populations
- —clinician interaction patterns
As a result, its strengths, limitations, and safety boundaries only emerge through sustained use in real clinical environments. Short pilots and paper benchmarks are insufficient.
This reality fundamentally shapes both procurement strategy and the role of startups.
---
The Role of Startups: Learning Systems, Not Finished Solutions
Startups building clinical AI do not deliver a final solution upfront.
They deliver learning systems that evolve through feedback and iteration.
Their value lies in:
- —rapid adaptation
- —tight clinician feedback loops
- —early exposure of edge cases and failure modes
This is precisely what early-stage clinical AI needs — and precisely what classical procurement struggles to accommodate.
---
Switzerland: Trust-Building Outside EU Sovereignty
Switzerland operates outside EU sovereignty, even though its procurement laws are closely aligned with EU principles.
This has important consequences:
- —Public hospitals are directly exposed to national and cantonal legal scrutiny
- —Procurement is treated as a mechanism of institutional trust and taxpayer protection
- —The tolerance for ambiguity is low
As a result, Swiss hospitals tend to:
- —separate innovation from procurement
- —frame early AI work as research or evaluation
- —delay tenders until uncertainty is significantly reduced
In practice, Switzerland uses early phases to build trust:
- —trust in safety
- —trust in robustness
- —trust in clinical appropriateness
This makes Switzerland exceptionally strong at legitimizing clinical AI, but structurally weak at scaling it early.
---
The Netherlands: Solution-Building Within EU Sovereignty
The Netherlands operates fully within EU sovereignty and the EU procurement framework.
Crucially, EU law:
- —allows proportionality
- —explicitly supports innovation partnerships
- —permits staged and negotiated procedures below thresholds
Dutch hospitals actively use this flexibility.
As a result, procurement is treated less as a legal risk surface and more as a strategic instrument to:
- —shape emerging markets
- —grow capabilities over time
- —move from pilots to long-term partnerships
In the Netherlands, the emphasis is on building solutions, even while uncertainty remains.
---
Trust vs. Solutions — A Structural, Not Cultural Difference
This contrast is not about competence or ambition.
It is largely structural:
- —Outside EU sovereignty (Switzerland)
- —higher sensitivity to legal challenge
- —stronger separation of innovation and procurement
- —trust must be earned before commitment
- —Inside EU sovereignty (Netherlands)
- —more room for proportionality
- —procurement used as an innovation tool
- —solutions can be built while learning
Both models are rational responses to their legal environments.
---
What This Means for Startups and Hospitals
The most effective strategy is not to force convergence — but to use each system for what it is good at.
- —Switzerland excels at:
- —rigorous evaluation
- —safety validation
- —clinical legitimacy
- —The Netherlands excels at:
- —scaling validated systems
- —operational deployment
- —sustainable commercialization
Startups benefit from engaging with both:
- —Swiss rigor makes their systems credible
- —Dutch pragmatism makes them viable
---
Procurement Should Follow Understanding, Not Replace It
Public procurement remains essential under any sovereignty regime.
But for clinical AI, procurement should be the result of learning, not a shortcut around it.
Hospitals that first invest in structured evaluation — in partnership with startups — make better, safer, and more future-proof procurement decisions.
---
*Clinical AI is not bought into existence.
It is learned into the clinic — and only then, responsibly procured.