There’s a growing disconnect between the promises being made about artificial intelligence in health care and what’s being delivered. If you’re a health system leader, it’s time to take a discerning view—not because AI lacks potential, but because many of the solutions entering the market are more marketing than medicine.

We’ve seen this dynamic before.

Earlier this month, a fintech founder was charged with defrauding investors after promoting an AI-powered shopping app as revolutionary. The pitch? One-click checkout from any website, powered by proprietary AI. The reality? A team of humans in a call center in the Philippines manually processed transactions. The automation rate? Nearly zero.

It’s a cautionary tale for health care. Too many AI vendors are showcasing sleek designs without the underlying capability. They claim to be autonomous systems, when in fact, they rely on basic automation—or maybe even manual labor—behind the scenes.

Meanwhile, Rock Health reports that over the past five years more than $27 billion has been poured into AI-enabled digital health startups. As AI rises to the top of investor priorities, the health care market is being flooded with solutions—many of which offer little more than automation overlays or clinician-facing dashboards. The result is a crowded and confusing landscape where it’s increasingly difficult to distinguish proven, scalable tools from speculative or superficial offerings.

As a physician, I’ve seen firsthand that successful technology in health care must do more than analyze data or surface insights—it must orchestrate action between the care team and the care receiver. That’s why we spent more than a decade building foundational infrastructure to support the real work of primary care delivery before layering in artificial intelligence.

Our platform now supports care for more than 8 million patients, including 1.5 million in value-based arrangements, with over $10 billion in medical expenses under management. These are not pilot programs or prototypes. They are scaled, operational systems delivering measurable outcomes.

Among our most advanced partners, we’ve helped achieve total cost-of-care reductions of 20 to 25 percent. And perhaps most importantly, 98 percent of patients in these partnerships see their primary care physician each year. That level of engagement is not driven by a chatbot or interface—it’s the result of coordinated systems working behind the scenes to close gaps and extend care.

Tom, our Primary Care as a Service (PCaaS) solution, integrates data from 63 electronic health record pipelines and 75 payer feeds. With tens of billions of clinical and consumer data points normalized and in play, Tom enables intelligent automation based on the aggregated medical record and predicted health trajectory of individual patients to support them and their care teams while executing timely interventions.

We’ve focused on solving the hard, structural problems—data quality, system connectivity and workflow integration—because without those elements, AI becomes just another layer of complexity. It may look impressive, but it doesn’t meaningfully change how care is delivered or experienced.

Health system leaders should ask pointed questions: How is the technology improving patient access, care coordination or cost management? Can the vendor provide verified performance data and real-world use cases? What health care experience is informing their development? How does the implementation not just replace existing workflows but think about a new way of care delivery? Solutions that cannot demonstrate meaningful, measurable impact should prompt closer scrutiny.

Artificial intelligence holds tremendous potential to reshape health care—but only when it is grounded in real infrastructure, informed by clinical expertise and designed to operate within the complexity of care delivery.

We didn’t start with AI. We started with health care. Then we built the systems to support it. Only after that did we introduce intelligent tools to extend the reach of the care team and help close the primary care gap that stands at 2 billion hours annually in the U.S. alone.

The real work isn’t in the interface. It’s in everything behind it.

David Carmouche is a physician executive.