Appeals Are 0.45% of the Money
Analysis · Pinnacle Services Corporation · September 2026. Figures are Recovr model output on the twelve-month denial book of a national collection of more than 12,000 physicians; see the notes on method at the end.
Almost every product sold to recover denied claims is organized around the appeal. The category is called denial and appeal management. The demo is an appeal letter. The metric is an overturn rate. It is a reasonable place to start if the question is which denials are the hardest to win. It is the wrong place to start if the question is where the money is.
Over the twelve months to September 2026, a national collection of more than 12,000 physicians in more than sixty specialties, billing 699 payers, recorded $40.2 million in denied charges on 279,575 claims, coded under 163 distinct claim adjustment reason codes. Recovr's models identified $14.74 million of that as recoverable. Of the recoverable amount, the categories where an appeal is the correct instrument — medical necessity, coverage policy, prior authorization and timely filing — account for $65,715. That is 0.45%.
The other 96.8% sits in categories where the right action is a corrected claim, a reroute to the correct payer, or a refile. Not an argument. A resubmission.
Two axes, not one
Denial taxonomies go wrong by conflating two questions. The first is why the payer denied, which is what the CARC and RARC codes encode: eligibility, coding, wrong payer, request for information, unprocessable, policy, timely filing. The second is what recovers the money, and there the list is short: correct and refile, provide documentation, appeal, reroute, verify eligibility, obtain authorization, bill the patient, or write it off.
The mapping between the two is many-to-one and not intuitive. CARC 16, the unprocessable-claim code, is not appealable at all: CMS attaches remark code N704, "you may not appeal this decision," and it appears on every such remit in the book. CARC 109, wrong payer, is a routing problem. CARC 18, duplicate, is a reconciliation problem. Only medical necessity, some coding denials and some prior-authorization denials are genuinely appeal territory. A product organized around appeals is organized around the action that applies to the smallest slice of the book.
Where the recoverable dollars sit
Category | CARC codes | Denied | Claims | Recoverable | Share of recoverable | Action that recovers it |
|---|---|---|---|---|---|---|
Duplicate | 18 | $10,657,414 | 59,907 | $10,270,771 | 69.7% | Reconcile, refile |
Unprocessable, missing information | 16 | $2,621,223 | 36,047 | $1,588,402 | 10.8% | Correct and refile |
Wrong payer, misroute | 109, B11 | $1,404,338 | 14,599 | $1,312,538 | 8.9% | Reroute |
Coding, bundling, modifier | 97, 4, 236, 5, 11, 181 | $3,121,064 | 49,625 | $819,375 | 5.6% | Correct and refile |
Documentation requested | 251, 252, 226, 164 | $15,112,766 | 10,995 | $247,325 | 1.7% | Provide documentation |
Coordination of benefits | 22, 23, 24 | $1,420,882 | 26,226 | $154,492 | 1.0% | Secondary claim |
Provider credentialing | 183, B7, 147 | $268,993 | 2,669 | $125,922 | 0.9% | Correct and refile |
Medical necessity, policy | 96, 50, 55, 167, 216, B15, 246 | $1,370,760 | 16,831 | $65,605 | 0.45% | Appeal |
Eligibility, coverage | 31, 242, 119, 27, 243 | $656,430 | 9,933 | $3,732 | 0.03% | Verify, reroute |
Prior authorization | 197 | $318,674 | 3,437 | $110 | 0.001% | Retro-authorization |
Timely filing | 29 | $1,051,219 | 16,313 | $0 | 0% | Appeal with proof |
Three things in that table deserve to be read slowly.
Pure medical necessity recovers nothing. CARC 50, the classic "not deemed a medical necessity," appears on 2,930 claims and $98,142 of denied charges, of which the models score $0 as recoverable. CARC 167, diagnosis not covered; CARC 216, review-organization finding; CARC B15, qualifying service not received — all zero. The only meaningful recovery in the policy bucket is CARC 96, non-covered charges, at $65,568, and most of that is a benefit-plan routing question rather than a clinical argument.
Timely filing is a cliff, not a category. $1.05 million denied on 16,313 claims, zero recoverable, all of it classified as avoidable. Once the clock runs, the money is gone; the leverage is entirely in the acknowledgment trail that proves the claim was submitted on time. That is a records problem, and it argues for keeping clearinghouse acknowledgments for as long as any payer's filing limit runs, not for a better appeal letter.
Prior authorization is a prevention problem wearing a recovery costume. $318,674 denied on 3,437 claims; 97.9% of it is classified avoidable and $110 is recoverable. The money is won by obtaining the authorization before the service, not by contesting its absence afterward. Any post-adjudication platform that claims prior authorization as a recovery opportunity is describing a category it cannot touch.
The money, in other words, is in the unglamorous codes: duplicates, missing patient identifiers, provider-record mismatches, modifiers, misrouted claims. CARC 16 alone carries 22,412 recoverable claims, the largest recoverable claim count in the book, at a modeled 94.4% success rate for the provider-mismatch variant. CARC 183, the ordering-provider eligibility edit, recovers at 99.1%, the highest rate of any code present. Neither is contested. Both are corrected.
What this means for how recovery gets staffed
The practical consequence is that appeal capacity is the wrong capacity to build first. An appeal is a clinical argument with a letter attached: it needs records, a policy citation and a reviewer. A corrected claim is a data operation: find the field, fix it, resubmit as a frequency-7 replacement, and confirm the clearinghouse accepted it. The second is faster, cheaper and, on this evidence, addresses two hundred times the dollars. A team that hires appeal writers before it automates corrections has staffed for the 0.45%.
It also explains a pattern anyone who has run a billing office will recognize. The medical-necessity queue is the one people talk about because it is the one that feels like work. The duplicate queue is the one nobody talks about because it feels like noise. In this book the noise is $10.3 million.
Notes on method, and what would change the result
These figures are model output, not realized cash. Every dollar labeled recoverable was identified as such by Recovr's recoverability models; the honest phrasing is "identified as recoverable," which is what the platform measures at this stage, and recoveries realized will be lower.
The duplicate category carries the result. CARC 18 is 69.7% of recoverable dollars at a modeled 89% success rate across 6,601 claims. A genuine duplicate is a write-off; a wrongly flagged one — common with panels, bilateral procedures and repeat same-day testing — is recoverable. If the model is optimistic about which is which, every percentage in this analysis moves. The appeal share would rise, but from 0.45% to something still well under 2%.
One payer artifact has been set aside. Of the $15.1 million in documentation-request denials, $14.47 million — 99.9% — comes from a single Medi-Cal managed-care plan on 3,052 claims at roughly $4,742 each. That is one plan's adjudication behavior, not a market phenomenon, and it is 36% of the gross denominator. Excluding it, the book is $25.7 million denied, of which 57% was identified as recoverable.
The book is one consolidated collection, but not one practice's. It spans more than sixty physician specialties, led by gastroenterology, general surgery, obstetrics and gynecology, internal medicine and dermatology, with roughly 8,800 physicians in the top twelve specialties and more than 12,000 in all. It is a physician book, not a laboratory one, and its payer mix is weighted toward California Medi-Cal managed care; a book weighted differently would move the category shares. The category buckets are our own mapping of 163 CARC codes to eleven reasons, stated in the table so that it can be disputed.
None of that changes the shape of the finding. Appeals are where denial recovery is marketed. Corrections are where it is earned.
Pinnacle Services Corporation builds Recovr, an AI post-adjudication claim-recovery platform for laboratories, physician groups and revenue-cycle companies. The category analysis above can be reproduced on any 835/837 file set.
