Insufficient documentation leads Medicare's $28.8B improper-payment bill. It's the coder's fix.

Two numbers from early 2026 point at the same gap. CMS reported that Medicare fee-for-service made $28.83 billion in improper payments in fiscal year 2025, a 6.55% error rate, and it said the leading cause was insufficient documentation, not fraud. Weeks later, a Medical Group Management Association poll of 288 practices named denials and appeals their single biggest revenue leak. One is the government clawing money back after the fact. The other is practices losing it in real time. Both come down to the same thing: the note didn't prove the code.

The cause CMS keeps naming is documentation, not fraud

The FY2025 improper payments fact sheet puts the Medicare FFS improper payment rate at 6.55%, or $28.83 billion, down from the 7.66% CMS reported for FY2024. It's the ninth straight year the figure has stayed under the 10% statutory threshold.

The line that matters for coders is the cause. CMS writes that most of those improper payments happened where a reviewer couldn't tell whether a payment was proper because of insufficient documentation, and it adds that insufficient documentation is generally not a sign of fraud or abuse. The fact sheet lists the error buckets plainly: no documentation, insufficient documentation, and documentation that doesn't substantiate the payment. The number itself comes out of CMS's Comprehensive Error Rate Testing program, which samples paid claims and re-reviews the record behind each one. Insufficient documentation means the record that arrived didn't prove the service. It doesn't mean the service was fraudulent, or even wrong.

Denials are now the biggest leak in the building

On January 6, 2026, MGMA asked practices where they lose the most revenue. Denials and appeals came first at 48%, ahead of front-end issues at 23%, billing and collections at 14%, coding at 13%, and charge posting at 2%, across 288 responses. MGMA's respondents described denials that "felt more automated and less negotiable," with more payer edits and the same recurring reasons: medical necessity, invalid authorization, timely filing.

Strip the labels and most of those denials are documentation arguments. A medical-necessity denial says the record didn't connect the service to a condition that justified it. A "claim lacks information" denial says a required element was missing. You can appeal, but the appeal wins or loses on what the note already said on the date of service, and by then the note is frozen.

Why "insufficient" almost never means "blank"

The trap in the word insufficient is that it sounds like a missing note. It rarely is. The encounter happened, the provider documented something, the coder picked a defensible code. What's missing is the link a reviewer needs: the diagnosis stated to the specificity the record supports, the condition shown as evaluated or treated on that date, the medical necessity for the service spelled out instead of implied.

Risk-adjustment coders already work to that standard. An HCC isn't supported because a diagnosis sits on a problem list. It's supported when the encounter shows the condition was monitored, evaluated, assessed, or treated. A CERT reviewer on a fee-for-service claim and a RADV reviewer on a risk-adjustment chart are asking a version of the same question: does the record prove what the claim asserts. The MEAT criteria hold both to the same bar.

What a claim needs to survive the review

The fix isn't "document more." It's documenting the one element the reviewer will look for. This table maps CMS's own improper-payment cause categories to the gap in the record and the desk-level fix.

None of that is work invented for an audit. It's the same record behind a clean claim and a survivable appeal, and you build it once.

The pre-bill habit that beats the appeal

An appeal is the most expensive place to fix a documentation gap. The record is closed, and the staff time goes to arguing instead of preventing. MGMA's write-up points at the front end for the same reason: build financial clearance, meaning eligibility, coverage, and auth status, before the date of service, and lean on checklists for the service lines that draw the most denials. On the denials themselves, its advice is to appeal what matters and go after the root cause, not to fight every line.

For a small practice, the coding version of that is a short, boring pre-bill pass on the service lines that draw denials: check that each billed diagnosis is coded to the specificity the note supports, and that the reason for the service is on the page. Line up the supporting elements in an evidence worksheet so an appeal, if it comes, arrives as one packet instead of a scavenger hunt. When the note is ambiguous about whether a condition was actually addressed, that's a query, not a guess, and our provider query templates cover phrasing that stays compliant.

Where the human coder is still the control

Payers are automating the denial side, and MGMA's respondents felt it as edits that are harder to argue with. Practices are getting pitched documentation and coding tools to keep up. Whatever a tool drafts, the element a reviewer demands is still a judgment call: is this diagnosis supported to this specificity, does this note show medical necessity, is this the code the record proves. A tool can surface the gap. It can't decide the record proves the claim. The coder signs off that it does. Confirm any code you're about to defend against its official description in the Code Book, and check the codes on a specific encounter in the encoder while you work the note.

What this is and is not

None of this says CMS found $28.83 billion in fraud. CMS is explicit that insufficient documentation generally isn't fraud. The point is narrower and more useful: the largest preventable category of Medicare payment error, and the biggest revenue leak practices report, are the same gap seen from two sides. What CMS recovers and what a practice loses to denials both come down to whether the note proves the code. That's not a billing-department problem. It's the coder's.