Our team spent three days at HLTH Europe this year, and if there was one theme that came up in almost every conversation, it wasn’t which AI model someone was using. It was what happens after the model works.
Company after company told a version of the same story: they’d built something that worked — a diagnostic algorithm, a triage model, a movement-analysis engine — and the hard part wasn’t the AI anymore. It was everything sitting around it.
Four Things Stood Out
The Model Is Rarely The Bottleneck
One founder described their core algorithm as their “secret sauce” — fully built, fully theirs, not going anywhere. But the same team was stuck on something much less glamorous: getting their system to talk to different imaging servers across hospitals. Another team had a working AI agent but no clean way to plug it into the workflows and data their customers actually had. The pattern repeated all week: the model was done. The data, the integration, the governance around it — that’s where projects stall.
Teams Are Drawing A Hard Line Around Their “Real” IP
Almost nobody wanted help with their core AI or algorithm — that stays in-house, tightly held. What they wanted help with was everything adjacent to it: configuration tooling, dashboards, integration layers, the parts of the product that let a hospital or clinician actually use the AI day to day. AI implementation, in practice, is turning out to be mostly an engineering and product problem wrapped around a small AI core.
Value Is Being Measured In Outcomes, Not Novelty
The teams with the most confidence weren’t the ones with the flashiest model — they were the ones who could point to something concrete: fewer unplanned admissions, faster triage, earlier detection. AI-as-a-feature is losing ground to AI-as-a-measurable-outcome, and that’s a healthier place for the industry to be.
AI Is Starting To Point Inward, Not Just At Patients
Maybe the most interesting shift: several teams talked about using AI to simplify their own operations — making software configuration easier for internal teams, partners, and customer IT staff, not just improving a patient-facing feature. AI-for-the-product-team is becoming as real a use case as AI-for-the-patient.
What This Means For European Healthcare Specifically
Europe adds a layer that’s easy to underestimate from the outside: fragmentation. A hospital in Germany, a clinic in Poland, and a health system in the UK aren’t just different customers — they’re often running different imaging systems, different EHR setups, different compliance regimes (GDPR, MDR, national data residency rules). An AI model that works in a demo has to survive that fragmentation before it means anything clinically. That’s less a technology problem than an integration and compliance problem, and it’s exactly why so many of the conversations at HLTH kept circling back to “how do we actually get this running everywhere we need it to,” rather than “does the AI work.”
Where This Lines Up With What We Do
This is the layer Empeek spends most of its time in — not building the AI model itself, but the engineering underneath it: getting a working algorithm connected to real hospital data, real imaging systems, real EHR and PACS environments, in a way that holds up under compliance requirements across different markets. If your team has the AI working and is running into the same wall — the data layer, the integration layer, the “how do we actually deploy this across different environments” layer — that’s a conversation we’re always happy to have.
Loading...
