Payers are masking denials and underpayments as contractual adjustments, code reclassifications, and modifier changes that never appear in denial work queues — silently eroding practice revenue in increments too small to investigate individually but collectively representing hundreds of thousands of dollars in annual losses. AI revenue integrity detection compares expected versus actual reimbursement on every claim, exposing the hidden payment reductions that no human billing team would ever catch.
This isn't speculation. At HFMA's September 2026 executive roundtable — a closed-door session with seven health system revenue cycle leaders sponsored by Solventum — the pattern was described in explicit, operational terms by people running multi-billion-dollar revenue cycles.
"Payers are getting creative. They're masking things and not sending them through as denials; they're sending them through as a contractual discount, or they'll tell us, 'We changed the code, and we paid you less once we changed the code.' Those types of things are not going to come through as a denial. As a result, we miss some of the really bad behavior. And we don't have technology today that can assist us with that."
— Ashley Teeters, formerly VP of Revenue Cycle at TMC Health, speaking at the HFMA September 2026 roundtable.
That last sentence is the one that matters: "We don't have technology today that can assist us with that." She's right — most practices don't. But AI revenue integrity systems built specifically for this problem do exist. And they're about to become the dividing line between practices that protect their revenue and practices that bleed it invisibly.
The Anatomy of a Masked Denial
Traditional denial management operates on a simple principle: when a payer rejects a claim, it issues a denial with an ANSI reason code (CO-4, CO-16, CO-197, etc.). That denial enters a work queue. A billing specialist reviews it, determines appeal viability, and either reworks or writes off the claim.
Masked denials bypass this entire system. Here's how they work:
- Code reclassification: Your practice submits CPT 99214 (established patient, moderate complexity). The payer silently reclassifies it to 99213 (low complexity) and pays the lower rate. The difference — typically $40–$80 per visit — is posted as a contractual adjustment, not a denial. No denial code. No work queue entry. No appeal window triggered.
- Bundling edits applied post-submission: The payer bundles two separately billable procedures into one, reduces the total payment, and posts the reduction as a standard contractual adjustment. The bundling edit may not align with your payer contract, but unless someone manually compares the ERA against the fee schedule for that specific procedure combination, it looks routine.
- Modifier stripping: Modifier-25 (significant, separately identifiable E/M service) is quietly removed from the claim. The E/M component is denied or reduced, but the reduction appears in the ERA as a contractual adjustment rather than a modifier-specific denial.
- Retroactive rate changes: The payer adjusts contracted rates for specific codes without formal notification and begins paying at the lower rate. Each individual underpayment is small — $15, $30, $50 — but across hundreds of claims per month, the cumulative impact is massive.
In every case, the claim is technically "paid." It shows up in your paid claims report. Your billing company reports a clean claim. Your denial rate looks fine. But you're collecting less than you're owed, and nobody knows.
Why This Problem Is Getting Worse in 2026
Three forces are converging to accelerate masked denials:
1. Payer AI is scaling denial volume beyond human capacity
The US Senate Permanent Subcommittee on Investigations documented in October 2024 that payers are using AI to issue denials at rates 16 times higher than typical — specifically in Medicare Advantage post-acute care. That AI doesn't just generate formal denials. It also powers the automated code review, modifier auditing, and bundling logic that produces masked payment reductions at scale.
When a payer's AI can review every claim and apply payment-reducing logic automatically, the volume of subtle underpayments increases exponentially. No billing team can manually audit every ERA line item against every contracted rate for every payer.
2. Write-off thresholds are dropping
Thea Campbell, Solventum's Global Business Director for Revenue Cycle, noted at the HFMA roundtable that denial write-off thresholds have dropped from $750 to $50 across the industry in just five years. Payers aren't stealing in large, obvious amounts anymore. They're chipping away at revenue in increments small enough that the cost of manual investigation exceeds the individual underpayment. A $30 downcoding on a 99213 isn't worth 20 minutes of staff time to research. But 500 of those per month is $15,000 in silent annual revenue loss — from a single payer on a single code.
3. The reactive-proactive gap persists
The HFMA roundtable revealed that nearly half of health system leaders report their revenue cycle is still evenly split between reactive manual processes and proactive AI automation — a 50/50 ratio that hasn't shifted materially in two years. That means half of all revenue cycle work is still reactive: responding to denials after they happen, working queues that only contain formally denied claims, and missing the entire category of revenue loss that never enters a queue at all.
What Traditional Denial Management Cannot See
The fundamental limitation isn't capability — it's architecture. Traditional denial management systems are designed to monitor denial codes. They watch for CO, PR, and OA group codes with specific reason codes that indicate claim rejection. They're excellent at what they do.
But masked denials don't produce denial codes. They produce payment variances — differences between what should have been paid and what was actually paid, buried inside ERA adjustment lines that look identical to legitimate contractual adjustments.
To detect masked denials, a system needs to do something fundamentally different from denial management:
- Calculate expected reimbursement for every claim, at the line-item level, based on contracted rates, fee schedules, payer-specific rules, and modifier logic
- Compare expected versus actual payment on every remittance, flagging variances that exceed a configurable threshold
- Distinguish legitimate adjustments from masked reductions by cross-referencing adjustment reason codes, payer contract terms, and historical payment patterns
- Aggregate pattern data across thousands of claims to identify systematic payer behaviors — such as consistently downcoding specific CPT codes for specific providers
No human billing team can perform this calculation on every claim. The math alone — expected reimbursement at the line-item level across multiple payers with different fee schedules — requires computational capacity that exceeds what a spreadsheet or manual review can deliver. This is precisely what AI agents built for revenue integrity are designed to do.
How AI Revenue Integrity Detection Works
AI revenue integrity operates on a different plane than denial management. Instead of monitoring what payers formally reject, it monitors what payers actually pay against what they should pay.
Layer 1: Expected Reimbursement Engine
The system maintains a continuously updated model of every payer contract, fee schedule, and payment rule. When a claim is submitted, the AI calculates the expected reimbursement at the line-item level — including procedure-specific rates, modifier impacts, place-of-service adjustments, and applicable bundling rules. This isn't a static lookup table. It's a dynamic model that learns from actual payment patterns and adjusts as payer behavior changes.
Layer 2: Payment Variance Detection
When the ERA arrives, the AI compares actual payment against expected reimbursement for every line item. Variances are categorized: legitimate contractual adjustments (matching the contracted rate), patient responsibility adjustments (deductible, copay, coinsurance), and unexplained variances — amounts that don't align with any contractual term, fee schedule entry, or known payer rule.
Layer 3: Pattern Intelligence
Individual variances might be noise. But when the AI detects that a specific payer is consistently paying $45 less than expected on 99214 visits for a specific provider, or systematically applying bundling edits to a specific procedure pair that the contract doesn't support, the pattern becomes actionable intelligence. This layer transforms individual data points into strategic denial recovery opportunities worth pursuing in aggregate.
Layer 4: Automated Recovery Workflow
Once a masked denial pattern is identified, the system generates an appeal or dispute with the supporting documentation: the contracted rate, the expected reimbursement calculation, the actual payment, and the variance. For patterns affecting dozens or hundreds of claims, batch recovery actions can recoup tens of thousands of dollars from a single identified behavior. Ashley Teeters noted at the HFMA roundtable that AI generates completely rewritten second-level appeal letters in 30 seconds versus 1 hour manually — turning bulk recovery from a staffing problem into an automation problem.
The Revenue Impact Practices Are Missing
The scale of masked denials is difficult to quantify precisely because the defining characteristic is invisibility. But the data points from the HFMA roundtable paint a clear picture of the financial exposure:
- 16× denial volume increase from payer AI (Senate investigation) means the raw volume of payment-reducing actions — both formal and masked — is orders of magnitude higher than five years ago
- $750 → $50 write-off thresholds (Solventum) means payers are engineering underpayments to fall below investigation thresholds, maximizing the total revenue they can claw back without triggering human review
- 50% of revenue cycles still reactive (HFMA roundtable) means half the industry has no mechanism to detect payment behavior that doesn't produce a formal denial code
Conservative industry estimates suggest masked denials and hidden underpayments represent 1–3% of net revenue at practices relying solely on traditional denial management. For a practice collecting $5 million annually, that's $50,000–$150,000 in silent, ongoing revenue loss. For a health system collecting $500 million, the exposure is $5–$15 million.
Carol Plato, VP of Revenue Cycle at North Mississippi Health Services, highlighted a related gap at the roundtable: real-time eligibility (RTE) vendors "aren't providing enough depth" — telling practices "this person has insurance" but missing hospice status, SNF status, and other denial-triggering factors. The information gap extends from the front end to the back end. If your insurance verification misses coverage nuances and your payment analysis misses masked reductions, you're losing revenue on both sides of the claim lifecycle.
What Smart Practices Are Doing Now
The HFMA roundtable participants offered several practical frameworks for addressing masked denials:
Move from denial management to revenue integrity
Denial management asks: What did the payer reject? Revenue integrity asks: Did the payer pay correctly? The second question catches everything the first question misses. Practices deploying AI revenue integrity systems alongside traditional denial management create a complete revenue protection layer that addresses both visible and invisible revenue loss.
Implement hard stops, not warning stops
Danielle Reese, VP of Patient Access at Hackensack Meridian Health, described replacing warning stops with hard stops in Epic — ensuring that upstream data quality issues are caught before claims are submitted, reducing the surface area for payer adjustment exploitation. Centralized financial clearance and QA teams provide proactive denial prevention at the front end.
Use short-term AI contracts
Joseph Koons, SVP and Chief Revenue Officer at LifeBridge Health, recommended short-term AI contracts because the technology is evolving too rapidly for long-term commitments. This is practical advice: the AI tools available for masked denial detection today will be materially more capable in 12 months. Lock in outcomes, not vendors.
Reclassify when the data demands it
Jonathan Davis, Executive Director of Revenue Cycle at Yale New Haven Health, shared a telling example: when 90% of specific inpatient cases lost on appeal, his team implemented policy changes to reclassify those cases as outpatient from the start — eliminating the denial entirely rather than fighting it repeatedly. The same logic applies to masked denials: once AI identifies a systematic payer behavior, the practice can adjust coding or submission processes upstream to prevent the reduction from occurring.
The Strategic Imperative
The HFMA roundtable concluded with a statement that frames the urgency: "With AI, payers can issue denials almost instantaneously and sometimes at rates 16 times higher than typical. This shift has turned AI adoption from an elective efficiency tool into a strategic necessity for protecting health system revenue."
That's the framing shift. AI for hospital revenue cycle isn't a nice-to-have productivity tool anymore. It's defensive infrastructure. And the specific gap that needs closing isn't formal denial management — most practices have that covered. The gap is revenue integrity: detecting the payment reductions that never enter a denial queue, never trigger an alert, and never get worked.
Payers have no incentive to make masked denials visible. They have every incentive to make them smaller, more frequent, and harder to detect. The only entity that benefits from catching them is the practice. And the only way to catch them at scale is AI that compares expected versus actual reimbursement on every single claim, every single day, for every single payer.
The practices that deploy this capability first will recover revenue they didn't know they were losing. The practices that wait will continue bleeding — invisibly, silently, and permanently. The choice isn't about technology adoption. It's about whether you're willing to let payers define what counts as a denial and what doesn't.