The prior authorization debate is focused on the wrong variable.
Turnaround time mandates and gold-carding address the payer side of the equation. The provider side — the operational, financial, and coverage cost absorbed by health systems when PA decisions go wrong — remains largely unmeasured. You cannot close an asymmetry you have not quantified.
The asymmetry nobody is naming
Insurers began deploying AI in utilization review years before most health systems had a strategy for responding to it. The result is a structural asymmetry — payers making adverse decisions faster, at higher volume, with algorithmic consistency, while providers track the impact in disconnected systems that never produce a single number leadership can act on.
The scale is not subtle. Medicare Advantage insurers made nearly 53 million prior authorization determinations in 2024, according to KFF. The AMA’s 2025 physician survey found that practices now complete an average of 39 prior authorizations per physician per week, with physicians and staff spending 13 hours weekly on PA-related work. Ninety-three percent of physicians report that prior authorization delays access to necessary care. Twenty-nine percent report that it has led to a serious adverse event for a patient.
When denials are appealed, they are overturned at rates that raise fundamental questions about the quality of the initial decision. A 2022 HHS OIG investigation found that 75 percent of appealed Medicare Advantage prior authorization denials were overturned at the first level of review. A 2026 OIG report on skilled nursing facility admissions found that MAOs overturned 95 percent of appealed denials in favor of the enrollee. KFF similarly found in 2025 that 82 percent of appealed MA denials were successful.
This is not a technology gap. It is a measurement gap that technology can close — if health systems decide to close it.
Maryland named it publicly — and that matters
In 2025, Maryland’s Governor and General Assembly took a step most states have not. They formally acknowledged that adverse decisions in the state health care system were rising — and they mandated a response.
The Maryland General Assembly passed HB 995/SB 776, directing the Maryland Insurance Administration and the Health Service Cost Review Commission to jointly staff a workgroup to study the rise in adverse decisions across the state health care system. That workgroup has held six public meetings since November 2025, with a seventh scheduled August 20, 2026. Every session has been recorded and every presentation published.
What Meeting 6 — held July 30, 2026 — examined is particularly significant: the use of artificial intelligence in utilization review denials. That conversation is now on the public record in Maryland. It confirms what providers have suspected but struggled to document — that the tools making adverse decisions are becoming more sophisticated while the tools measuring their impact are not keeping pace.
Meeting 7 is August 20, 2026 and is open to the public. Registration is available here.
What health systems are not measuring — and what it is costing them
PA cost at most health systems is captured at the point of denial — physician time, appeal cycles, claim delays. The downstream cost is not captured at all. That downstream cost includes:
Staff hours spent on follow-up, resubmission, and appeal cycles across clinical and administrative teams. A 2026 AJMC systematic review found that completing a single prescription drug prior authorization can take up to 64 minutes and cost providers between $15 and $63. In pharmacy settings, the mean time is 24 minutes — 16 for pharmacists, 8 for technicians. Multiply that across 39 PAs per physician per week and the labor cost is substantial.
Prescriptions on PA hold that expire before approval. Patients who abandon before the PA resolves — particularly in safety-net populations with transportation barriers and coverage instability. 340B eligible fills that never happen because the PA cycle outlasts the patient’s next appointment. Coverage lapses that begin with a PA delay and end with a lost patient.
None of these appear on a standard report. They live in separate systems — pharmacy, eligibility, revenue cycle, clinical — that are never connected into a single PA impact view. The total cost stays invisible to the executives who would act on it if they could see it.
The FQHC dimension — where the asymmetry hits hardest
For safety-net providers, the payer-provider asymmetry in PA is not just an administrative problem. It is a patient care and financial sustainability problem simultaneously.
A commercially insured patient denied on PA has options — follow up, appeal, switch providers. A Medicaid patient at an FQHC operating under work requirements, coverage redetermination pressure, and transportation constraints may not return at all. That failed PA becomes a lost patient, a lost 340B eligible transaction, and an unreported revenue impact — all from a single adverse decision made by an algorithm the provider never saw.
Maryland’s workgroup is examining this at the policy level. The operational response has to happen at the health system level — and it starts with measurement.
What AI enables on the provider side
If insurers are using AI to make adverse decisions faster and at greater scale — the logical response is not more staff. It is better data, organized and surfaced in a way that enables health systems to understand PA impact across all departments simultaneously, respond to it strategically, and advocate against it with documented evidence.
Integrated PA impact data enables four things:
Payer negotiation with real numbers. A health system that can present documented staff time, abandonment rates, 340B transaction loss, and coverage lapse impact tied to a specific payer’s PA decisions has a fundamentally different contract conversation than one asserting a general burden.
Targeted internal investment. When PA cost is measured by cycle stage — submission, follow-up, appeal, abandonment — leadership can see exactly where the process is failing. The answer is rarely more people. It is usually a specific broken handoff that no one has quantified before.
Provider-side voice in policy. Maryland’s workgroup has payer data. What it does not yet have is granular provider-side operational cost data in a form that speaks to the full system impact. Health systems that build this measurement become the credible voice in that policy room — not with anecdotes, but with documented numbers.
A justified AI deployment target. Payers have already decided that AI belongs in utilization review. Health systems that cannot measure PA impact across departments cannot build or justify an AI-assisted response. The data is the foundation. Without it, the response stays reactive.
The data infrastructure to measure PA impact across departments is not a future investment. For health systems absorbing AI-assisted adverse decisions at increasing volume, it is a current operational gap with a measurable cost.
Sources
2025 AMA Prior Authorization Physician Survey
KFF — Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024
AJMC — Prescription Drug Prior Authorization: Costs to Pharmacies and Physicians (2026)