Two-Thirds of Your Denied Claims Never Get Reworked. That Is Revenue You Already Earned.

A denied claim is not a lost claim. It is revenue you already earned, sitting in a queue your billing staff does not have the hours to work. When two-thirds of those claims are never reworked, the write-off is a staffing decision wearing a payer's clothing.

I have come to believe this is the single most misread problem in healthcare revenue cycle management. Practice administrators look at a 10 or 11 percent denial rate and think they have a payer problem. The payer is a factor, but the real rupture is downstream: a billing team that receives 100 denials a week and, under realistic staffing conditions, can actively appeal maybe 35 of them. The other 65 never get reworked before the timely-filing window closes, at which point they either age into write-off, get partially adjusted, or get reposted to patient responsibility.

The AMA and Change Healthcare put that never-reworked rate at roughly 65 percent of all denials. That is not a statistic about payer aggression. That is a statistic about queue capacity.

Initial denial rate at intake, 2024

The 2024 Change Healthcare Denials Index put the average initial claim denial rate at approximately 11.8 percent, up from 10.2 percent in prior benchmarks. MGMA sets the target floor at under 10 percent, with a clean-claims rate above 95 percent. Practices running at 11.8 percent are already outside the benchmark band on the first submission. That gap compounds: every re-submission cycle adds operational days to the accounts receivable aging, and most commercial payer contracts apply timely filing limits in the 90 to 180 day range from date of service (Medicare's outer limit is one calendar year). Miss that window and the claim becomes genuinely unrecoverable, not just difficult.

Experian Health's 2025 State of Claims survey, drawn from 250 RCM leaders, found that 41 percent report at least one in every ten claims denied. The top cited causes are missing or inaccurate data, authorization gaps, and inaccurate patient information at registration. These are upstream errors producing downstream denials. A billing team reworking claims on the back end is, in part, paying for process failures at the front desk.

Triage by CARC and RARC code

The Claim Adjustment Reason Code (CARC) and Remittance Advice Remark Code (RARC) system is the denial's native language. Every electronic remittance advice (ERA) from a payer arrives with these codes. CARC 97 means the benefit is included in the allowance for another service. CARC 4 means the service dates are inconsistent. CARC 50 is a non-covered service. Each code points to a specific appeal pathway, a specific attachment requirement, a specific regulatory argument.

A competent billing specialist reads those codes, pulls the clinical notes, drafts the appeal letter with the appropriate CPT and ICD-10 substantiation, and submits through the payer portal. In my client practices, that work runs roughly 20 to 45 minutes per claim depending on complexity, which is consistent with MGMA workflow estimates in the same range. With 65 denials abandoned per 100 received, that is roughly 1,300 to 2,925 person-minutes of work not happening each week in a mid-sized practice.

An AI agent integrated into the practice management system (athenahealth, Tebra (formerly Kareo), AdvancedMD, or Cerner Millennium on the hospital side) can parse the ERA, classify the denial by CARC/RARC, match it against prior adjudication patterns, and produce a draft appeal with the relevant documentation checklist. A billing staff member reviews, attaches the clinical record, and submits. That review step takes 5 to 8 minutes in the deployments I have run.

The agent does not draft the clinical record. It drafts the argument structure and the attachment list, which is where most of the time was going.

The break-even denial value

This is the operator point most administrators miss. MGMA's administrative cost benchmark for denial rework sits at approximately $25 per claim for a standard re-submission. Change Healthcare's broader estimate, including complex appeals with clinical documentation, runs up to $118 per claim. Other industry ranges go as high as $181 for multi-touch denials requiring peer-to-peer physician review. The AHA reported that providers spent approximately $19.7 billion appealing denials in 2022. That is not an appeals budget. That is a tax on operational complexity.

The cost-per-rework figure matters for a different reason than most administrators use it. Practices cite it when arguing for more billing staff. The more precise use is to calculate the threshold claim value below which rework is economically irrational at current labor costs. If rework costs $25 to $118 and the denied claim is worth $80, the math argues for write-off.

In the FloridAI deployments I have direct visibility into, an AI-assisted workflow reduces the per-claim staff-touch cost to an estimated $8 to $12 of reviewer time. I am presenting that as an operator estimate from a small sample, not as a vendor benchmark. At that touch cost, claims that were economically irrational to rework become economically rational to rework. That is a win-win for throughput and margin.

The 86 percent avoidability problem

Change Healthcare's Denials Index also found that approximately 86 percent of denials are potentially avoidable. Contrast that with the 24 percent of denials the same index classifies as unrecoverable once submitted. The gap between those two numbers is the operational target: the majority of denials should never occur, and of those that do occur, the majority can be recovered if worked in time.

The avoidability finding points back to registration and authorization workflows. Prior authorization denials, which map heavily to CARC 15 (the authorization number is missing, invalid, or does not apply) and CARC 197 (precertification absent), are generating claims that will be denied at adjudication regardless of clinical accuracy. Practice management systems like athenahealth and AdvancedMD already flag missing authorizations at scheduling. What they do not do is generate the payer-specific prior authorization packet, work the Availity submission (Availity is the dominant clearinghouse hub in Florida for this), and track the determination back to the appointment. That is the gap the agent fills. The PM flags; the agent works.

The downstream fix addresses what reaches the queue anyway. For that segment, the agent's value is throughput: working a much larger share of the denial queue without adding headcount.

Payer-side AI is the counterforce

A provider-side AI agent does not enter a static system. Payers have been deploying their own claims-review automation for years. ProPublica's reporting on Cigna's PXDX system documented batches of claims denied in seconds without individual physician review. The UnitedHealth/naviHealth litigation alleges algorithmic post-acute denials that strip Medicare Advantage coverage. The structural reality is that provider-side AI improves throughput inside an adversarial system where payer-side AI is simultaneously increasing denial volume and reducing appeal acceptance latency.

That changes the framing. Throughput on the appeal side is necessary but not sufficient. First-appeal success rates may compress over time as payer automation hardens. The defensible position is to work more of the queue, faster, with better documentation, while pushing the upstream registration and authorization workflows hard enough that fewer denials enter the queue in the first place. The agent is not a recovery tool. It is a queue-management tool inside a market that is getting more adversarial, not less.

Appeal letter drafting and submission queue

The mechanics of the appeal vary by payer and by denial type. For Florida Medicare, Part A and Part B are administered by First Coast Service Options (Jurisdiction N), not Palmetto GBA; DME claims fall under CGS Administrators (Jurisdiction C). Medicare standard redetermination must be requested within 120 days of receipt of the initial determination notice per 42 CFR 405.942. Commercial payers like UnitedHealthcare, Cigna, and Aetna have their own portals and their own appeal timelines, defined in the payer contract. Florida Statute 641.3155 sets the prompt-pay framework for HMO claims (clean claims paid within 20 days for electronic submission, contested claims resolved within defined windows), which is the statute Florida operators actually need to know for managed care timing disputes.

An agent that maps the CARC code to the payer-specific appeal pathway, drafts the appeal letter with the claim number, service dates, CPT codes, and initial denial rationale, and populates the payer portal submission form is doing the lookup and drafting work that currently stands between a denial and a submission. The billing specialist's role shifts from drafting to reviewing and certifying. That shift is how you increase appeal volume without increasing headcount.

What I have observed across the small set of FloridAI client deployments: appeal submission rates moving from the 35 to 40 percent range of the denial queue into the 80 to 90 percent range within 60 days of deployment, and first-appeal collection rates in the 40 to 60 percent band, which industry sources (HFMA, MGMA) place in roughly the same range as a sector benchmark. I am presenting this as operator observation with a small n, not as a published study. The math on recovered revenue at those rates is substantial for any practice running north of $3 million in annual collections.

Before you evaluate a vendor in this space

Four questions an operator should put directly to any vendor offering AI-assisted denial management:

1. Does your system parse ERA files natively and classify by CARC and RARC code, or does a human still have to categorize the denial before the workflow starts? If a human categorizes first, you have not changed the throughput constraint, you have just added a software layer on top of it.

2. What is the per-claim staff-touch time in your reference implementations, and can you show me the before and after numbers from a practice in the same specialty and payer mix as mine? Florida payer mix is not Ohio payer mix; Medicaid managed care penetration here is different, and authorization requirements vary by plan.

3. How does your system handle timely filing deadlines across payers? Missing a 90-day window on a $400 claim because the agent queued it incorrectly creates liability, and I need to know where that tracking lives and who owns the error.

4. What percentage of the denial queue in your reference practices is reaching submission, what is the first-appeal success rate by CARC category, and how is that success rate trending as payer-side automation tightens? Aggregate numbers obscure which denial types the system handles well, which it does not, and whether the recovery rate is holding up against the payers most likely to deny by algorithm.

Those four questions will tell you whether a vendor has built a real operational workflow or a demo layer on top of a large language model. The difference is in the specifics, and the specifics are where revenue either recovers or ages out.