Dwight Johnson, director of revenue cycle at Southeast Health in Dothan, Alabama, has three priorities for 2026. All three involve AI. Number one: payer automation for prior authorizations. Number two: predictive models that flag likely denials before claims are filed. Number three: automated appeals with minimal human intervention.
He is not innovating. He is responding. Because on the other side of every claim, the payer is already running the same playbook — in reverse.
The payer’s head start
Insurance companies were early AI adopters in the revenue cycle, and they pointed it in one direction: denial. Machine learning models scan incoming claims for coding discrepancies, medical-necessity gaps, and authorization defects — at a scale and speed no human review team can match. Down-coding algorithms shave reimbursement line by line. Prior authorization requirements, already the fastest-growing denial category, are administered through automated portals with rejection rates that providers describe as capricious.
The results show in the data. The HFMA/Guidehouse 2026 survey found 88% of providers say claim disagreements are preventing them from getting paid. Twenty percent report denial rates above 5% of claims, up from 12% in the prior survey. Seventy-four percent report increased prior authorization delays. Hospitals lost an estimated $48.4 billion to final denials and uncollected balances in 2025 — a 25% increase over 2024, per TipRanks’ analysis of industry data.
The payer AI advantage compounds because payers see every provider’s claims. A model trained on millions of submissions learns exactly which coding patterns, which diagnosis-procedure combinations, which documentation gaps predict a successful denial. The provider, seeing only its own claims, trains on a fraction of the data. It is an asymmetric information war, and the side with more data is winning.
The provider counterattack
Hospitals are now deploying what the industry calls “counter-AI” — and the capabilities arriving in 2026 are genuinely formidable.
Predictive denial prevention. Before a claim is submitted, AI scans it against payer-specific denial patterns and flags likely rejections. The best systems push first-pass clean claim rates above 96%. The model knows that Payer X denies this procedure code without that modifier, because it has seen Payer X do it a thousand times.
Autonomous coding. Traditional computer-assisted coding is being replaced by agentic AI that analyzes clinical charts, extracts CPT and ICD-10 codes, and submits claims without human touch for over 70% of standard cases. Coding lag — the days between discharge and claim submission — collapses to minutes. Every day of lag is a day of accounts receivable; the working-capital math alone justifies the investment.
AI-drafted appeals. When denials land, agentic systems draft appeal letters citing clinical guidelines and precedents, processing thousands in the time a human handles one. Joyful Health, which expanded its AI denial platform into full lifecycle RCM in October 2026, reports processing over $1.4 billion in denial-related transactions with a 6x return on recovered claims and resolution rates above 95%.
Payer intelligence. Hospitals are grading their payers — tracking denial rates, appeal overturn rates, and payment timeliness by insurer, then using the data as leverage in contract negotiations. An insurer that denies 12% of clean claims does not get the same rate discussion as one that denies 3%.
The vendor cavalry arrived in force this year. Oracle Health unveiled five AI revenue-cycle tools in September 2026 — prior authorization automation, clinical documentation integrity, charge capture, professional fee coding, and denial prediction — embedded natively in its EHR. The EHR vendors realized what the point solutions knew: the denial fight is won or lost inside the clinical workflow, not in a bolt-on dashboard.
The uncomfortable lag
Here is the number that should worry every hospital CFO: just 2% of providers say they have fully or mostly integrated AI across revenue cycle operations, per the HFMA/Guidehouse survey. Sixty percent have not implemented any AI or automation in the revenue cycle at all. Thirty-nine percent are running point solutions — a prior-auth bot here, a coding assistant there, nothing connected.
The gap between the 2% and the 60% is the competitive story of the next three years. The hospitals in the 2% are pulling away on every metric that matters: days in A/R, clean claim rate, denial overturn rate, cost to collect. The hospitals in the 60% are fighting payer AI with human labor — bringing spreadsheets to an algorithmic war.
Becker’s “Ask a Revenue Cycle Leader” series captures the urgency. Lynn Ansley at Moffitt Cancer Center is leading an Epic implementation explicitly to transform revenue cycle performance. Joel Gentry at OU Health is prioritizing AI-driven prior authorization workflows for the second half of 2026. The language has shifted from “exploring AI” to “deploying AI” — but deployment remains the exception.
The counterargument: the arms race framing is incomplete
The AI-vs.-AI narrative is compelling but risks obscuring a simpler truth: most denials are still preventable without any AI at all. Experian’s data shows missing or inaccurate claim data drives 50% of denials, authorizations 35%, registration errors 32%. These are front-end process failures — eligibility verification, documentation completeness, timely filing — that better staffing and workflow discipline would fix. An AI denial predictor is valuable; a registrar who verifies eligibility correctly the first time is more valuable.
There is also the question of whether the arms race is socially productive. Payer AI that denies valid claims and provider AI that appeals them is, in aggregate, an expensive zero-sum game — billions in technology spending to move dollars back and forth across the same table. The HFMA’s own framing acknowledges this: executives want stronger payer relationships, not just better weapons. The optimal end state is fewer denials, not better-fought denials.
And the technology has failure modes. AI coding agents trained on historical data can systematize historical errors. Predictive models can create false confidence — a 96% clean-claim rate still leaves 4%, and the 4% clusters in the complex cases where human judgment matters most. Automation without oversight is how a misconfigured rule denies a thousand claims before anyone notices.
Where this ends
The arms race does not end with one side winning. It ends when the economics of fighting exceed the economics of preventing — when both payers and providers realize that the billions spent on denial and appeal automation would be better spent on clean submission and straight-through processing. Value-based care, for all its slow progress, points in this direction: when payment is tied to outcomes rather than claims, the denial as a profit lever disappears.
Until then, the hospitals that thrive will be the ones that field both weapons and diplomacy: AI sharp enough to win the fights worth fighting, process discipline strong enough to prevent the fights that should never happen, and payer relationships honest enough to resolve what technology cannot. The 60% still running on human labor do not have three years to catch up. The payers are not waiting.
HospitalDenial.com is the domain for the discipline hospitals can’t afford to treat as a back office. Currently available — inquiries to info@primedomaincapital.com.