When Denial Rates Jump From 8.7% to 22.7%

The figure comes from a Senate subcommittee report, not a peer-reviewed journal. But it is the best public quantification we have of what happens when a payer rolls out a predictive denial engine at scale: UnitedHealth’s post-acute services denial rate climbed from 8.7% in 2019 to 22.7% in 2022, right alongside the deployment of its tool nH Predict. The report alleges that UnitedHealthcare, Humana, and CVS use predictive technology to systematically deny Medicare Advantage patients access to post-acute care. I cannot independently verify the methodology, and the source is politically charged, but the magnitude — a 2.6× increase in three years — is hard to dismiss as noise.

Providence CFO Greg Hoffman told Healthcare Dive that underpayments and initial denials rose more than 50% over two years as payers adopted AI claims review tools. Providence had to increase human touches per claim by over 50% just to keep up with the new documentation requirements. The hospital system entered a 10-year partnership with R1 RCM partly to fight back. When a large nonprofit system — not known for extravagant spending — commits to a decade-long vendor deal, it signals that the denial problem has moved from operational nuisance to strategic threat.

The Asymmetric Economics of Denial

Payer AI deployment is not about faster processing. It is about automated prior-authorization checks and predictive models that flag claims for denial before a human ever looks at them. nH Predict, Blue Shield’s Salesforce-based prior-auth tool, and similar systems are designed to surface patterns that justify a denial — often technicalities that would have been missed by manual review. The incentive structure is straightforward: every denied claim is money the payer does not have to pay. The Senate subcommittee report described it as “systematic denial,” but the technology itself is neutral; what matters is how it is used.

The economics favor the payer: invest once in an AI model and generate savings on every denied claim. Providers, by contrast, must invest in appeal software and additional staff to contest each denial — with no guarantee of recovery. The same AI that costs a payer pennies to run can cost a provider dozens of dollars to fight.

How many providers are actually fighting back? A 2025 survey by Experian, cited by Cedar, found that only 14% of providers use AI to reduce denials. I hesitate to treat 14% as a precise measurement — Experian’s methodology and sample are not fully disclosed in the source — but the directional gap is consistent with every other survey I have seen. The same Cedar article notes that 69% of those who do use AI report fewer denials, which is promising but self-selected. The 14% figure is still the most commonly cited benchmark, even if it is a secondary citation.

A more detailed picture comes from an HFMA survey (February 2026, n=95): 27% of organizations are actively deploying AI at scale across multiple RCM functions, 53% are conducting pilots. The sample is small — 95 respondents — so treat the distribution as directional. The key insight is that even among early adopters, the bulk of deployment is still in pilot phase, not full production. The arms race is real, but the provider response is still in the early stages, and most of the $18 billion spent overturning denials in 2025 was human labor, not AI.

The $18 Billion Tab and the 60% That Never Gets Paid

The financial toll is where the conversation moves from interesting to urgent. The American Hospital Association estimates hospitals spent nearly $18 billion overturning denials in 2025. Premier, a membership organization representing roughly 4,400 U.S. hospitals and health systems, reports that the average cost to work a denied claim is $57.23, up from $43.84 the prior year. Premier's data comes from its member base, which is large but not necessarily representative of all providers; the $57.23 figure likely reflects larger systems with more complicated claim portfolios.

The most striking number is from McKinsey, reported through secondary sources (Moveo and HFMA): 60% of denied claims are never appealed because the manual investigation cost exceeds the recovered value. Think about that. For six out of ten denials, the provider simply writes off the revenue. It is cheaper to lose the money than to fight for it. This is the structural problem that AI could address — automated appeal generation and prioritization — but only if the investment makes economic sense for the provider.

Why Collaboration Won't Fix This

There is a persistent hope that payers and providers will eventually collaborate to reduce unnecessary denials. The Oliver Wyman survey of over 200 decision-makers found that 92% agree there are “no-regret” AI investments to pursue. But Isaac Sieling of Huron put it bluntly: “The fundamental disconnect between payers and providers is not a technology challenge.” It is a financial incentive mismatch. Payers save money by denying; providers spend money to recover. No amount of AI changes that calculus.

“The fundamental disconnect between payers and providers is not a technology challenge.” — Isaac Sieling, Huron, cited in HFMA

The same Oliver Wyman survey indicates 20–40% of organizations report broad enterprise-wide use of AI in RCM, and 70–90% expect to increase spending. Those ranges are wide because adoption varies enormously by health system size and region. The sample of >200 is small for an industry-wide claim, but the direction is consistent: spending is rising on both sides. The arms race is not going to de-escalate. The question for CFOs is not how to stop it, but how to ensure their organization is not the one spending more to fight than the denied revenue is worth.

What CFOs Should Expect Next: Talent Gaps and BPO Shifts

William Blair estimates that AI could save nearly $20 billion annually in the RCM space for providers — but that is under an assumption of full adoption. The same report notes that about 80% of survey respondents cite lack of AI talent as the single largest barrier to implementation. The near-term savings will be far lower. The provider RCM software market is roughly $25 billion today, expected to grow to $40 billion by 2030, according to William Blair. Meanwhile, 80% of total RCM spend ($80 billion) currently goes to business process outsourcing organizations. The shift from BPO labor to software is the real opportunity — but it will take years, and it requires the talent that most organizations do not have.

For health system CFOs and revenue cycle executives, the takeaway is not that AI will fix the denial problem next quarter. It is that the asymmetric dynamic is structurally baked into the current reimbursement system. Payers have a direct financial incentive to deny claims at scale, and they have shown they are willing to deploy AI aggressively to do it. Providers must invest in counter-AI not because it is a competitive advantage, but because the cost of staying manual is already unsustainable — and getting worse. The 60% of denials that go uncontested represent billions in lost revenue that no cost-cutting initiative can recover. Whether AI can meaningfully lower that number remains to be seen, but the alternative, doing nothing, is already costing more.