"The people sending the denials have more time, more tooling, and more patience than the people answering them. That is not a complaint. It is the business model." That's John Beene, founder of Soupy Audit, writing in Healthcare IT Today in August 2026. It's the most honest description of the healthcare revenue cycle anyone has published this year — and it explains why provider-side AI isn't optional anymore. The denial economy is asymmetric by design. Pretending otherwise keeps providers losing slowly.
Payers run continuous, instrumented processes to identify, deny, and recover payments. Providers respond case-by-case, under deadline pressure, with staff whose primary job is something else. The numbers make the asymmetry concrete: denied claims cost hospitals nearly $20 billion per year (HFMA/Waystar 2026), 55% of providers say claim errors are increasing (up from 44% in 2022), and the Medicare FFS improper payment pool stands at $31 billion — the reservoir RAC extrapolations draw from. The only way to close this gap is to build the same kind of continuous, instrumented process on the provider side. That's what AI does.
The Asymmetry Is the Business Model
The denial economy doesn't exist because of mistakes. It exists because the economics work for the side that initiates denials and recoveries. Payers invest in denial infrastructure — teams, algorithms, audit contractors — because every dollar denied or clawed back is a dollar retained. Their processes are continuous, automated, and patient. They don't operate under filing deadlines. They don't run out of staff at 5 PM. They don't have to choose between answering the phone and working an appeal.
Providers operate under the opposite conditions. Billing staff manage denial responses between registration duties, patient calls, and prior authorization workflows. Appeals require clinical documentation that clinicians don't have time to write. Filing deadlines create hard ceilings that payers don't face. And the cognitive burden of treating each denial as a unique, one-off problem means the systemic patterns driving denials go unrecognized.
This is why 65% of denials are never appealed — not because providers agree with the denial, but because the operational cost of responding exceeds the perceived recovery value on a case-by-case basis. At scale, that calculus bleeds hundreds of thousands of dollars annually from mid-size practices and millions from health systems.
RAC Extrapolation: How 30 Claims Become a Seven-Figure Demand
Recovery Audit Contractors operate on a simple and devastatingly effective model. They review a sample — typically 30 to 50 claims — calculate an error rate, and then extrapolate that rate across the provider's entire claim universe. A six-figure sample becomes a seven-figure demand letter.
The math is straightforward. If a RAC reviews 40 claims and finds a 15% error rate, they apply that 15% across every claim in the defined universe. A provider with 10,000 claims in the universe and an average reimbursement of $500 faces an extrapolated demand of $750,000 — derived from a sample of 40 claims.
Here's what almost no provider does: audit the auditor's math.
CMS Medicare Program Integrity Manual Chapter 8 requires specific procedural safeguards for valid extrapolation:
- Clearly defined universe — the claim population the sample is drawn from must be precisely specified
- Documented sample-size justification — the sample must be statistically valid for the universe
- Stratification executed as declared — if the auditor said they'd stratify by service type, they must have actually done it
- RAT-STATS seed preserved for reproducibility — the random seed used for sample selection must be documented so the sample can be reproduced
Miss any one of these procedural requirements and the extrapolation is challengeable on procedure alone — before anyone even argues the clinical merits. Yet almost no one on the provider side checks. The letter arrives, the number is large, and the provider's instinct is to negotiate the total rather than challenge the methodology.
The defensible point of demand is the one-sided 90% lower confidence bound recomputed from the provider's actual defense outcomes — not the auditor's opening number. The gap between those two numbers is where settlement lives.
This is a math problem before it's a legal problem. AI treats it as one — analyzing every extrapolated demand for procedural compliance, recalculating confidence bounds based on actual defense outcomes, and identifying the defensible settlement range before a single dollar is paid.
The Upstream Leak: Surgical Capacity Evaporating from Pending PA
The denial economy's most invisible cost isn't in the denials themselves — it's in the revenue that never generates a claim at all.
Consider the scenario Beene describes: A surgeon blocks an OR for Tuesday. The patient's prior authorization is still pending Monday at 5 PM. The block goes unfilled. The patient gets rescheduled. The peer-to-peer review never gets scheduled. The case is done at a competitor three weeks later — or doesn't happen at all.
No claim was denied. No letter was sent. Capacity just quietly evaporated.
This upstream leak is invisible to traditional denial management because there's no denial event to manage. The revenue loss doesn't appear in denial reports, appeal win rates, or rework queues. It shows up in OR utilization rates that drop by 10-15%, in procedure volumes that flatten despite growing patient demand, and in surgeons who spend Tuesday mornings doing office visits instead of operating.
For surgical specialties — ENT, orthopedics, ophthalmology — this capacity leakage can represent the single largest revenue loss in the practice, exceeding traditional denial costs by a factor of three or more. A $60,000 surgical case that never reaches the OR because PA was pending doesn't show up as a $60,000 denial. It shows up as nothing. An empty block. A rescheduled patient. A surgeon who had time to operate but nothing authorized to operate on.
The Fix: Continuous Instrumented Process on the Provider Side
The denial economy is not going to become more symmetric on its own. Payers have no incentive to level the playing field — the asymmetry is where the margin lives. The only path forward is to build the same kind of continuous, instrumented process on the provider side that already exists on the payer side.
What that means operationally:
1. Pre-Op Readiness That Flags Auth Risk Before the Block Is Held
AI monitors every scheduled procedure against its authorization status — not on the day before surgery, but continuously from the moment the case is booked. When a patient's PA shows signs of delay (payer response latency, missing documentation, pattern-matched payer behavior), the system flags it days in advance. The OR block is either defended with proactive peer-to-peer scheduling or released early enough to backfill with an authorized case.
This isn't denial management. This is capacity defense — preventing the upstream leak that traditional denial workflows never see.
2. Payer and Auditor Pattern Tracking
AI tracks which payers and auditors are running which plays this quarter. When a RAC increases review activity in a specific CPT range, AI alerts the practice before the first demand letter arrives. When a payer shifts denial patterns — new edit rules, changed medical necessity criteria, retrospective modifier audits — AI detects the pattern from early denials and adjusts pre-submission validation across the entire claim pipeline.
This is the vendor watch layer. Payers already track provider billing patterns in detail. AI gives providers the same intelligence about payer behavior.
3. Capacity Balance Surfacing
AI continuously surfaces OR minutes and appointment slots at risk from pending authorizations, documentation gaps, and payer-specific delay patterns. Practice leadership sees — in real time — exactly how much capacity is leaking, which payers are causing the leakage, and which cases need intervention to prevent the leak.
Without this visibility, the capacity loss is invisible until it shows up in quarterly revenue reports. With it, the practice can intervene at the case level, the payer level, and the scheduling level before revenue evaporates.
4. Math-First Clawback Defense
Every extrapolated demand is treated as a math problem first and a legal problem second. AI validates the procedural compliance of every extrapolation — was the universe properly defined, was the sample statistically valid, was stratification executed as declared, is the RAT-STATS seed preserved? Then it recomputes the defensible demand based on actual defense outcomes, identifying the gap between the auditor's opening number and the statistically defensible settlement point.
This approach has transformed RAC defense from an expensive legal exercise into a quantitative discipline. The math doesn't lie, and most extrapolations have procedural weaknesses that reduce the defensible demand by 30-60% before clinical arguments even begin.
| Defense Layer | Without AI (Reactive) | With AI (Continuous) |
|---|---|---|
| PA / Capacity Defense | Discover pending PA day-of → cancel/reschedule | Flag auth risk at booking → defend or backfill days ahead |
| Payer Pattern Intelligence | Notice trends after months of denials accumulate | Detect payer behavior shifts from early signals, adjust pipeline |
| Capacity Visibility | OR utilization reviewed quarterly, leakage invisible | Real-time dashboard of at-risk capacity by payer and case |
| RAC Extrapolation Response | Negotiate total; rarely challenge methodology | Validate procedural compliance, recompute defensible bound |
| Denial Appeals | 65% never appealed; staff bandwidth limits response | 100% coverage; same-day appeal generation on every denial |
The $60,000 Documentation Failure
HFMA's 2026 Mid-Revenue Cycle Roundtable documented a case that illustrates the asymmetry in miniature: an orthopedic surgeon using skin substitutes without documenting medical necessity, generating repeated denials at $60,000 to $70,000 each. The problem wasn't the procedure — it was the disconnect between physician compliance and coding documentation.
Diana O'Connor, VP of Clinical Services at Waystar, put it directly: "Accuracy of clinical documentation is vital to achieving a high-performing revenue cycle." Yet as Heather Wilson noted at the same roundtable: "It's 2025, and we're still operating with spreadsheets and emails and faxing information."
The siloed operations problem — physician compliance in one lane, coding documentation in another, eligibility verification in a third — is the structural vulnerability that payers exploit. Each silo operates independently, and the gaps between silos are where revenue leaks. AI closes the silos by connecting clinical documentation to coding validation to payer-specific submission rules in a single, continuous pipeline.
Nikki Harper from Mayo Clinic named the patient impact: practices have to ensure they're "not operating in silos" because the patient gets caught in the middle of disconnected billing processes. The operational fix and the patient experience fix are the same thing — integrated, continuous, instrumented processes that don't depend on staff manually bridging departmental gaps.
Healthcare AI Governance: Security and Continuous Awareness
The IO State of Information Security report finds that 47% of healthcare organizations identify AI-driven phishing as a significant threat, 51% cite AI-generated misinformation, and 55% experienced a third-party or supply chain security incident. Meanwhile, 51% report budget constraints affecting security initiatives and 47% identify skills shortages.
The governance parallel to the denial economy is direct: annual risk assessments are as insufficient for security as case-by-case denial responses are for revenue defense. AI-enabled platforms receive frequent updates, vendors introduce new capabilities, and data flows evolve as systems become more interconnected. Governance must become an operational function — continuous awareness, accountability, and resilience — not a compliance exercise performed once a year.
For providers deploying AI in revenue cycle operations, this means the AI itself must be governed with the same continuous instrumentation it provides. Audit trails on every decision, compliance monitoring on every payer interaction, and security architecture that treats interoperability as an attack surface, not just a feature.
The Pipeline Mindset
The denial economy is a pipeline on the payer side — measurable inputs, a measurable conversion rate, and a number at the end that responds to attention. Providers who continue treating each denial letter and each rescheduled surgical case as a one-off will continue losing to a system designed to beat one-off responses.
BAM AI builds the continuous, instrumented pipeline on the provider side: pre-op readiness that prevents capacity leakage, payer pattern intelligence that detects denial shifts before they compound, math-first clawback defense that challenges extrapolation methodology, and same-day appeal generation on every denial — not just the ones staff have time to work.
The opening is to stop pretending the denial economy is fair and start operating with the same pipeline sophistication the other side of the table already uses. AI makes that possible for a 10-physician practice, not just a 500-bed hospital.
Book a demo to see the continuous instrumented denial defense pipeline running on your claim data — and find out how much revenue you're losing to an asymmetry you haven't been measuring.