Healthcare AI: Why Don’t Clinicians Use It?

Why do we keep buying healthcare AI that nobody uses?

Health systems spent an estimated $45 billion on AI and analytics tools in 2025. That number is climbing in 2026.

And yet, when you walk into most clinical settings, the tools are either turned off, ignored, or buried three clicks deep in a workflow nobody follows.

I have seen this firsthand. Not just at small FQHCs, but at large integrated health systems with entire innovation teams.

The technology is not the problem.

Here is what I actually think is happening.

🔎 We are measuring adoption by contract signatures, not by clinical use.

A health system buys a predictive readmission tool. IT deploys it. The vendor sends a press release. Leadership presents it at a board meeting.

Six months later, the alert is firing into a workflow nobody checks. The model is right 70% of the time. Nobody is acting on it.

That is not a technology failure. That is a change management failure dressed up as an innovation win.

The second problem is alert fatigue.

Clinicians already navigate hundreds of interruptive alerts per shift inside their EHR. Most are ignored. When a new AI flag gets layered on top of an already overloaded system, it does not get more attention. It gets less.

We have not fixed the signal-to-noise problem. We have made it worse.

The third problem is trust.

Clinicians do not trust what they cannot explain. Most AI tools today output a score or a flag with minimal reasoning attached. When a physician does not understand why the model flagged a patient, the default behavior is to override it and move on.

Explainability is not a nice-to-have. It is the difference between a tool that changes outcomes and a tool that generates a report nobody reads.

Here is the question I keep asking health system and FQHC leaders:

Before you buy the next AI tool, can you tell me what happened to the last three you deployed? Not what the vendor said in the QBR. What actually happened at the point of care?

If the answer is unclear, the problem is not the AI.

The problem is the process you built around it.

We are in a moment where healthcare AI is genuinely capable of changing outcomes. Readmission prediction, sepsis alerts, care gap identification, risk stratification. The models work.

But a model that works and a model that gets used are two very different things.

I think the next frontier for health tech is not building smarter AI. It is building AI that actually fits into how clinicians work, in real time, without adding cognitive load.

That is a harder problem than the algorithm.

So here is the question I want to leave you with: If your health system had to audit every AI tool deployed in the last three years and measure actual clinical utilization, not licenses, not logins, but real workflow integration, what would you find?

👉 Follow me for daily healthcare insights and updates on LinkedIn.

🔍 For deeper analysis, subscribe to my Substack, where I share long-form articles, industry trends, and in-depth perspectives on healthcare, AI, diagnostics, and the future of care. → https://lnkd.in/eJKFuB_p