gshc2020.com

The difficult conversations almost never centered on software. They centered on trust. Reviewers wanted to know whether a recommendation reflected the patient’s situation or simply what the data happened to capture. Physicians questioned whether unusual cases would be treated fairly. Operations teams worried that efficiency gains could come at the expense of consistency or access.

Those conversations changed how I think about AI in healthcare. The technical side of building models is only part of the job. The harder responsibility is making sure technology doesn’t reinforce existing inequities or overlook the context that experienced clinicians consider every day.

If you’re leading technology, operations or clinical transformation initiatives, you’ve probably had similar discussions. In this article, I want to share a few lessons I’ve learned from working alongside clinicians and healthcare organizations about why human context remains essential, even as AI becomes a larger part of healthcare decision-making.

Why Human Context Matters

One of the biggest misconceptions surrounding AI in healthcare is the assumption that automation reduces the importance of clinical judgment. In reality, the opposite is occurring.

The most effective AI systems increasingly emerge from deep collaboration between data scientists, clinicians, utilization reviewers and operational teams. Predictive systems become more reliable when they incorporate tacit clinical reasoning rather than treating healthcare purely as a statistical exercise.

In our own work supporting prior authorization and medical review workflows, one recurring lesson has been that structured policies alone are insufficient. Real-world decisions depend heavily on contextual interpretation, longitudinal understanding and practical clinical nuance.

That’s why I no longer think of clinicians as end users of these systems. In my experience, they’re part of the design process.

Three Practices That Have Worked For Me

Over the past several years, I’ve noticed a few practices that consistently lead to better outcomes when organizations introduce AI into healthcare workflows.

First, involve frontline reviewers earlier than feels necessary. On one implementation, our initial design looked reasonable on paper, but experienced utilization reviewers immediately pointed out situations where patients with similar documentation required different decisions because of subtle clinical context. Those conversations changed how we approached the workflow long before anything reached production.

Second, make disagreement part of the process instead of treating it as a problem to eliminate. During policy reviews, I’ve encouraged clinicians, nurses and operational leaders to explain why they reached different conclusions on the same case. Those discussions often uncovered ambiguities in documentation or policy interpretation that would never have surfaced through technical testing alone.

Finally, continue reviewing difficult cases after implementation. One health plan I worked with regularly set aside complex cases for multidisciplinary discussion instead of assuming every decision should follow the same pattern. Those sessions frequently identified situations where operational processes rather than the underlying technology, needed to change. They also helped build confidence because reviewers could see that their expertise continued to shape the process.

None of these practices requires sophisticated technology. They require organizations to treat clinical expertise as an ongoing source of learning rather than a checkpoint at the end of a project.​

A More Nuanced Role In Fraud, Waste And Abuse

I see this challenge most clearly in fraud, waste and abuse programs because those decisions often have significant consequences for patients, providers and public resources.

Early in my career, I often heard people describe these efforts primarily as identifying statistical outliers. Over time, I’ve come to appreciate that unusual utilization patterns don’t always indicate inappropriate behavior. Sometimes they reflect legitimate clinical complexity that only becomes apparent when experienced reviewers examine the broader context.

That experience has reinforced my belief that healthcare organizations should avoid treating statistical signals as conclusions. They’re starting points for informed clinical review, not substitutes for it.

This matters because public health outcomes and program integrity are interconnected. Aggressive enforcement without sufficient clinical context can unintentionally create access barriers, while inadequate oversight threatens sustainability across public systems.​​

The Future Of Public Health Will Depend On Distribution

When I think about the future of public health, I don’t think first about algorithms. I think about whether organizations can apply new capabilities consistently and responsibly across the populations they serve.

Technology can help identify opportunities earlier, support more consistent decision-making and reduce administrative burden. But those benefits only matter if they reach the patients who need them most.

For me, that’s the real measure of progress. Success isn’t defined by how advanced a model becomes. It’s defined by whether healthcare organizations use technology in ways that improve care while preserving fairness, transparency and sound clinical judgment.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?