There is a familiar arc to healthcare AI initiatives. A promising tool is selected. A pilot launches with real enthusiasm. The early demo impresses everyone in the room. And then the project quietly fails to scale — and the post-mortem keeps surfacing the same culprit. It is rarely the model. It is the data feeding the model.
This is the uncomfortable gap at the center of healthcare AI in 2026. By some industry estimates, provider adoption of AI is on track to roughly double in just two years. Yet according to the Office of the National Coordinator, only about 43% of U.S. hospitals routinely engage in all four core domains of interoperability — finding, sending, receiving, and integrating electronic health information. We are racing to deploy data-hungry intelligence on top of a data foundation that, for most organizations, is still fragmented. The mismatch is the story.
AI is starving, and interoperability is the meal
Modern AI models are capable out of the box. What they cannot do is manufacture the clean, structured, longitudinal data they depend on. And in most health systems, that data is scattered — across multiple EHRs, claims platforms, lab and imaging systems, departmental databases, and a tangle of point-to-point interfaces that were each built for one purpose and never designed to feed analytics.
The resulting failures are predictable once you see the pattern. A predictive model trained on inconsistently coded claims produces unreliable predictions. A prior-authorization tool that pulls from an incomplete patient record misses the clinical context that would have changed its output. A coding assistant that performs beautifully at one facility may underperform at the next because the underlying data is structured differently. None of these are algorithm problems. They are interoperability problems wearing an AI costume.
The foundation has a name, and it is no longer optional
The good news is that the infrastructure to fix this has matured, and the regulatory momentum behind it has shifted from intention to action. Three layers matter.
Standardized data and APIs. FHIR has become the de facto standard for exchanging clinical data through APIs, and the United States Core Data for Interoperability (USCDI) defines a consistent baseline of what that data should contain. Together they turn “we have data somewhere” into “we can query a defined, structured dataset on demand” — which is exactly what an AI pipeline needs.
Integration and normalization. Standards alone do not clean your data. Between your source systems and any AI sits the unglamorous but decisive work of an interface engine and a normalization layer: translating legacy HL7v2 feeds, reconciling terminologies, matching patient identities, and resolving the duplicates and inconsistencies that would otherwise train your model on your own errors. This is where most data-readiness work actually happens.
Network-based exchange. The Trusted Exchange Framework and Common Agreement (TEFCA), the Qualified Health Information Networks (QHINs) at its core, and CMS’s emerging Interoperability Framework are collectively pushing the industry away from one-off, point-to-point integrations toward a scalable network-of-networks. For data and analytics leaders, that means access to a far more complete, longitudinal patient picture — the raw material that makes AI outputs trustworthy rather than brittle.
What this means for leadership
The strategic reframe is simple but consequential: interoperability is not a compliance chore that runs parallel to your AI ambitions. It is the precondition for them. The organizations that will get durable value from healthcare AI are not the ones buying the most models. They are the ones investing in the data and integration layer first, so that every model they deploy — this year’s and next year’s — has clean, complete, real-time data to work with.
That argues for a specific sequencing. Before the next AI procurement, audit where your clinical and administrative data actually lives and what state it is in. Map what it would take to expose it in a standardized, queryable form. Treat your interface engine, normalization logic, and exchange connectivity as strategic infrastructure, not back-office plumbing. And recognize the timing advantage: as TEFCA and the CMS framework scale, early, intentional adopters can turn interoperability into a competitive edge while slower movers are still treating it as paperwork.
The takeaway
Healthcare AI will keep getting more capable, and the temptation to lead with the model will only grow. But intelligence is downstream of data, and data is downstream of interoperability. Get the foundation right — standardized, integrated, normalized, exchangeable — and the AI you layer on top has a real chance of delivering. Skip it, and you will keep running pilots that dazzle in the demo and disappear in production. No interoperability, no intelligence. It really is that direct.
About Arinder Singh Suri
Arinder Singh Suri is the Founder and CEO of Taction Software, a healthcare IT and custom software development company specializing in EHR/EMR integration, HL7/FHIR and Mirth Connect implementations, and healthcare AI.
