When everyone uses AI for prior authorization, governance wins

Given the administrative burden associated with utilization review and dispute processes, it's no wonder physicians and other clinicians are turning to AI for help with the health care payment process. But when everyone's using it — not just physicians, but all stakeholders on every side of the process — its existence is no longer the difference-maker; governance is.

When I talk about physicians' role in this area, and why it's driving them toward AI, two key parts of payer-provider dynamics spring to mind. The first is prior authorization, which consumes an average of 13 hours of physician and staff time each week, according to the American Medical Association's 2025 survey on prior authorization. Across a year, that's equivalent to nearly 17 full-time workweeks — a huge resource commitment.

Prior authorization is nothing new, but the administrative burden it places on providers has undoubtedly climbed over the last few years, with direct consequences for patients. Regulators have taken notice. The Centers for Medicare & Medicaid Services' new rules require payers to issue decisions within seven days for standard requests and 72 hours for urgent ones. The requirement has been in place since January for Medicare Advantage, state Medicaid and federally facilitated exchanges under the Affordable Care Act.

Increasing caseload of payment disputes

The second part, health care payment disputes, makes that burden heavier still. To cite an example that has become the center of debate, disputes under the No Surprises Act have seen a dramatic increase over the past few years. The NSA was built with good intentions, protecting patients against unexpected medical bills when they receive emergency care or are treated by out-of-network providers. But the number of disputes filed under the NSA has run far past what the law's designers projected, with caseloads climbing continuously, CMS has reported.

As a result, in both the prior authorization and the NSA payer-provider dispute processes, physicians and their revenue cycle management vendors face mountains of document retrieval and evidence extraction, comparison and synthesis work. It's no surprise, given these circumstances, that AI is becoming commonplace.

AI across all sides of utilization review

This is the kind of information-processing work that this technology excels at. In fact, four in five physicians now use AI professionally, according to an AMA survey, with generative AI tools used to develop text, sift through records and forecast outcomes.

This is not an isolated trend. On the insurer side, 84% of health insurers now use AI or machine learning at some point during the review process, including utilization management and prior authorization, the National Association of Insurance Commissioners reports. And independent reviewers, the entities ruling on disputes, aren't immune either. While there is little data to show just how common AI use is among independent dispute resolution entities, I work with many directly and have seen firsthand how this shift is also playing out on the reviewer's side of the equation.

What happens when AI is table stakes?

If all sides of the process are using AI, as they are, then the marginal value of simply having the technology will naturally fall over time. The question, then, becomes "Who will use AI sensibly?" Differentiating one provider from another, and one physician's evidence from another's, will increasingly depend on how they deploy, supervise, audit and constrain the technology.

Once this shift happens — and I believe it is already underway — the most trustworthy organizations won't necessarily be the ones deploying AI the most aggressively. Instead, they'll be those with the discipline to know where AI genuinely does and doesn't add value and where human clinical judgment must stay central.

In our work, AI accelerates the evidence synthesis, surfacing the relevant clinical information from a 1,000-page record in minutes. But crucially, a human owns the decision. And that must remain the case for any physician compiling clinical records and providing further explanation within the utilization review or dispute resolution process.

Physicians need audit trails showing how AI-assisted evidence compilation has been conducted, and where, exactly, in the process AI is in action. That way, if AI makes an assertion, the physician should be able to move back through the chain and find the underlying source. That doesn't just make the evidence more effective and accurate; it also saves the physician time and headaches down the line.

AI must include guardrails

Some may argue that AI has no business operating anywhere near decision-making about which patients get care and who pays for it — a conversation that has become far louder following the launch of CMS' WISeR model and after the past few years of headlines about automated denials. Given all the reporting, I understand how they might have arrived at that stance, but the failures behind those headlines are the very reason we must move toward stronger guardrails.

Many documented failures have involved opaque criteria, poor oversight and incentives that rewarded speed over accuracy. Those are instances where governance has gone wrong, not AI use in and of itself, and they highlight the importance of rigorous, somewhat standardized compliance processes for AI use across the health care landscape.

We can't shut AI out altogether, even if we wanted to. If you give two practices or health systems the same health record but only allow one to use AI, the AI-augmented one will find the evidence gap that decides the case far more often. And this technology is transformative for physicians themselves, freeing them to spend more time with patients and get back to the clinical work they trained for.

Health care has been here before. When electronic health records arrived, adopting the technology was only part of the challenge. Building reliable and secure processes around it was an equally big problem.

AI use in payment disputes and utilization reviews is fast becoming a non-negotiable for all parts of the chain, including physicians. The real question is about building strong governance and guardrails that keep the technology on track and allow physicians and other clinicians to extract maximum value from it. The organizations that do will ultimately come out on top, securing the best, most accurate results and reducing their own massive administrative burden in the process.

Madhu Reddiboina is the founder and CEO of RediMinds, a health tech firm specializing in AI-assisted contested medical review.