Recovering missed billables with automated RVU audits

Methodology

How the result was measured

A fixed methodology, so the result reads as evidence rather than a headline: what was measured, over what period, how recovery is defined, how it was attributed, and the full governed-run counts.

Study and observation period

January 6 to June 30, 2026 (6-month engagement). Figures are the monthly steady-state average across the window.

Records reviewed

≈480 cardiology electrophysiology encounters per month (one physician’s billable volume). Each encounter is one governed reconciliation run.

Baseline methodology

A parallel manual audit of a stratified 120-encounter sample by the practice’s certified coding reviewer established the baseline capture rate and the corrected coding for each sampled encounter. OpenAdapt output was compared against that adjudicated baseline.

Definition of “recovered”

Recovered here means IDENTIFIED and queued: a recovery opportunity that OpenAdapt flagged and wrote to a review workbook and recovery email for the billing team to submit. It does NOT mean submitted, approved by a payer, or collected. Dollar figures are estimated recoverable value at the point of identification, not booked revenue.

Attribution method

An opportunity is attributed to OpenAdapt only where it flagged a differential the baseline sample confirmed and the billing team had not already queued. Results are reconciled against the existing worklist to exclude double-counting of opportunities that would have been caught anyway.

Manual effort, before and after

Before: roughly 6 physician and coder hours per month to run the audit by hand. After: roughly 1 hour per month, spent on exception review of halted runs and QA sampling. Net reduction of about 5 hours per month.

Application and surface

The practice’s EMR and RVU spreadsheets. GUI last-mile navigation and write, where no supported billing API reaches the step.

Deployment mode

Runs in the customer-controlled Windows environment, on the practice workstation over RDP as deployed. No PHI leaves the customer boundary.

OpenAdapt and pack versions

OpenAdapt Flow v1.23.0, RCM cardiology workflow pack v0.4.2.

Identity and effect contract

Each run binds to the patient and encounter identity, read out of band before any write. The effect contract asserts the specific coding change intended for that encounter, and nothing is treated as done until that asserted effect is confirmed against the persisted record.

Verifier implementation

Out-of-band read-back. After a write, the verifier re-reads the persisted field through an independent path and compares it against the asserted effect. Disagreement or ambiguity halts the run and routes it to a human. A run is never reported successful on an unverified write.

Governed-run counts

Every run, accounted for

Monthly steady-state counts. Verified, halted, and failed runs reconcile to the total. Verified rate 99.2%, at $0.03 per record. The number that matters most for a clinical write is the last one.

A halt is a designed outcome, not a failure: it means the verifier could not confirm the effect, so a human reviews it instead of a wrong value silently reaching the record.