30-Day Documentation & Revenue Lookback

We ran 624 of your visit notes through the Rafter engines. Here is what was already in your documentation.

Your clinicians are already doing the clinical work — treating heart failure, managing CKD, adjusting diabetes regimens. The gap is between what was done and what the documentation captures and defends. Over one month of visits we found 69 reviewable HCC findings (a further 18 low-confidence hits were suppressed by our confidence model and are not counted anywhere in this report), a visit-level coding pattern that under-codes 29% of billed visits, and 68 documented conditions that would not survive a RADV audit as written. Every finding below is tied to verbatim evidence from your own notes, and every number traces to a footnoted method in the appendix.

Framing note: this lookback sizes prospective opportunity — what the same clinical work would capture going forward with pre-visit documentation support. It is not a proposal to re-bill prior claims.

1Missed HCCs — evidence was in the chart, the code never landed

The CDI engine scanned each note's medications, labs, vitals, and narrative against 115 CMS-HCC V28 condition profiles, then scored every candidate through the same confidence model that gates our production physician queries. Findings the model wasn't confident in were suppressed, not shown. What remains is grouped by confidence lane:

Headline uses the high-confidence lane only ($138,910). Review-lane candidates ($67,119) require specialist judgment and are reported separately on purpose — we don't inflate the number we'd be measured against.

What a finding looks like — one patient, verbatim evidence

The query that would have been sent — before the visit

In production, this finding becomes a pre-visit query to the physician, rendered from our V28 template catalog and passed through a 7-point AHIMA/ACDIS compliance gate before it can be sent. This is the actual rendered output — note it contains no codes, no scores, and no financial language:

2E/M coding accuracy — your billed mix vs. what the documentation supports

Each note was independently leveled by the Rafter coding engine (2021 AMA MDM rules, catalog-based, evidence-linked per element) and compared against the code your team billed4. 602 of 624 charts had a billed E/M code available for comparison.

Established patient mix (531 visits) — billed vs. engine vs. CMS national reference5

By provider

Provider C leaves the most on the table; Provider D carries the most audit exposure. Same tool fixes both — that's the bidirectional point.

3MEAT evaluation — the HCCs you did document, and whether they'd survive an audit

Capturing an HCC is half the job; defending it is the other half. CMS RADV audits test whether each documented condition shows Monitoring, Evaluation, Assessment, or Treatment linked to that condition in the encounter. We assessed all 391 HCC-mapped conditions you documented this window with the Rafter MEAT engine (per-diagnosis, assertion-gated — a med-list mention alone does not count).

Completeness of MEAT evidence (0–4 elements per condition)

Top documentation patterns not meeting MEAT — all HCC conditions

In production, every one of these becomes a specific, one-line strengthening prompt at the point of care ("link the eGFR to your CKD assessment"), not a retrospective cleanup project. That's how the 82.6% moves toward 100% without adding documentation time.

3bMock RADV — the diagnoses you already billed, graded the way a RADV reviewer would

Section 3 started from your notes. This section starts from your claims. We took the 453 HCC diagnoses billed on 289 visits this window and asked the RADV question of each one: does the note for that visit support it? A patient's HCC counts as supported if any visit we reviewed supports it. Not graded: 21 billed HCCs outranked by a higher HCC billed for the same patient (they don't pay); 7 codes left for a coder to check.

Why billed HCCs didn't hold up

Largest unsupported billed HCCs

Billed codes in OIG's high-risk groups

OIG's Medicare Advantage audits found these groups unsupported most often (Toolkit A-07-23-01213, Dec 2023). Check each one against hospital, oncology or pharmacy records, whatever the note says.

4Billing cost comparison — what the same claims cost on Rafter

Your window volume annualizes to ~7,594 claims/yr7 against stated collections of $1.48M. Your current arrangement (~6% of collections, bundled) costs $88,800/yr. Rafter runs on a transparent stack — Stedi clearinghouse at ~$0.50/claim round-trip plus AI coding with near-zero marginal cost — so we can price below incumbents and show you the math:

5The offer — start where the money already is

Everything above was found in your charts by the same engines you'd be buying. The recommended path starts with the CDI platform (where this report found $220,242/yr) and adds coding and billing only when each step has proven itself.

Next step: a 30-minute walkthrough of this report with your physicians — we bring the per-patient evidence behind every number in Section 1, on screen, in the same dashboard your team would use.

Method, assumptions & disclaimers

How this audit was produced

638 charts received → 14 excluded for scan quality (listed separately for your records) → 624 analyzed. Each chart ran through the production Rafter engines — the identical code paths a subscribing practice runs: outpatient coding (11-phase, 2021 AMA MDM catalog engine; internally gated at 82% exact / 100% ±1 E/M agreement on validation corpora), outpatient CDI (9-phase CMS-HCC V28 scan, 115 condition profiles, confidence-calibrated), MEAT engine (per-diagnosis, assertion-gated), and the V28 query renderer (every sample query in this report passed the same 7-point AHIMA/ACDIS compliance gate used in production). 22 charts lacked a billed E/M code and are excluded from Section 2 accuracy denominators only. 18 low-confidence HCC hits were suppressed by the confidence model and excluded from all dollar figures.

Assumptions & footnotes

1. Annualized HCC dollars = HCC coefficient × $15,572.88 per patient-year — CMS's PY2027 national Medicare per-capita cost (FFS USPCC, aged + disabled: $1,297.74 a month × 12; CY2027 Rate Announcement, Table I-2) — applied to Medicare Advantage charts only, each on its own CMS-HCC segment, assuming continued MA enrollment and annual recapture. Charts on Traditional Medicare, commercial, Medicaid or an unconfirmed payer are counted, not priced.
2. Coefficients are the CMS-HCC V28 values for the segment each chart belongs to (community aged or disabled; dual status from the coverage on file, so a dual member is valued on the dual segment; rounded, illustrative in this sample). A chart with no confirmed payer lane shows the community, non-dual, aged value and is counted, not priced.
3. 70% validated-and-signed rate is a deliberately conservative planning assumption for physician confirmation of high-confidence findings; replaced by your measured rate after the first quarter.
4. E/M dollar deltas use 2026 Medicare PFS national non-facility rates (illustrative in this sample); commercial payer deltas are typically larger.
5. CMS national E/M mix reference: Medicare utilization data, family medicine specialty, most recent published year.
6. Collections-rate benchmark (85–90% in-house vs. 93–97% specialist-managed) is an industry reference range, not a Rafter guarantee.
7. Annualization multiplies 30-day window figures by 12.17. Claim count = encounters with ≥1 billable line.
8. Mock RADV (Section 3b): billed ICD-10 codes as supplied by the practice, mapped to CMS-HCC V28 (PY2026). Each billed payment HCC is graded against the note for the visit it was billed on; a patient's HCC counts as supported if any reviewed visit supports it, including through a more severe code in the same HCC family. The V28 hierarchy is applied across each patient's billed HCCs: an HCC a higher billed HCC outranks doesn't pay and isn't counted, and when a paying HCC isn't supported but a lower billed HCC it outranks is, only the difference counts. RAF at risk is the billed coefficient minus the coefficient of any lower-specificity code the note does support, valued at the same base rate as footnote 1. Visits outside the window were not reviewed, and a RADV reviewer may use any qualifying record from the payment year. RADV audits Medicare Advantage plans: Section 3b grades MA charts only; charts on other payer lanes are not graded.

Compliance framing

This lookback is opportunity sizing for prospective, going-forward capture. It is not an offer to re-bill, re-code, or amend previously submitted claims, and no prior-period correction is proposed here; any such review would be a separate, bidirectional exercise (adding supported and removing unsupported codes) conducted with your compliance counsel. Rafter pricing is never based on code volume — performance fees apply only to defensible, MEAT-validated, physician-signed documentation lift. Physicians never see financial figures inside queries; all sample queries herein contain no codes, risk scores, or dollar amounts, consistent with AHIMA/ACDIS 2022 compliant-query practice. Patient references are de-identified (initials, age/sex, date of service). Chart data was processed under BAA with document parsing performed locally; charts and derived data are purged per the data-handling agreement (default 90 days). This report contains estimates based on documentation review; actual results depend on payer mix, enrollment, and clinical documentation at the point of care.

Section 3b (Mock RADV) was run at the practice's request and reviews diagnoses on claims already submitted. It reports whether the note supports each one and proposes no corrections. Decisions about those claims belong to the practice and its compliance counsel.