Executive Summary
JIL Sovereign's Exposure Radar treats a set of compliance-attestation results as a proxy statistical review sample drawn from the paid-claim universe, then computes two figures from it: a point-estimate error rate, and a conservative one-sided 90% Wilson-score lower-bound error rate. Both figures are extrapolated onto the full paid-claim universe, segmented by branch, clinician, and payer, mirroring the statistical-sampling extrapolation methodology that a government or payer auditor would itself use to compute an overpayment demand.
Presenting both figures together - not just the single point estimate - gives the audited party a defensible best-case-to-worst-case range under the actual methodology it may eventually face, rather than a single number that a real audit could exceed.
Problem Statement
Statistical-sampling extrapolation - the practice of drawing a review sample, computing an error rate, and applying that rate to a much larger paid-claim population to compute an aggregate overpayment figure - is a codified methodology used in CMS and payer program-integrity audit guidance. It is opaque to the audited party until an actual audit happens: a provider or payer typically has no way to see, in advance, what number a real auditor's extrapolation would likely produce from its own current claims data, and therefore no way to prioritize remediation by projected dollar exposure before being audited.
Why Existing Approaches Are Insufficient
- Single point-estimate audit software: Reports only one error rate with no conservative counter-figure, understating how much worse a real extrapolation could plausibly land.
- Ad hoc, non-regulator-methodology risk scores: Not tied to the statistical extrapolation math an actual government or payer auditor would apply, so the resulting number is not directly comparable to a real audit's likely demand.
- No segmentation: Treating an entire organization as one risk pool, rather than by branch, clinician, or payer, gives no way to prioritize where remediation effort should go first.
Technical Architecture
| Step | Computation |
|---|---|
| 1 | Treat the set of compliance-attestation results as a proxy review sample of size n drawn from the paid-claim universe, with x observed errors. |
| 2 | Compute the point-estimate error rate x / n. |
| 3 | Compute a conservative one-sided 90% Wilson-score lower-bound error rate from the same (n, x) pair. |
| 4 | Extrapolate each rate against the dollar value of the paid-claim universe, segmented by branch, clinician, and payer. |
| 5 | Present the point-estimate and lower-bound extrapolated figures together, per segment. |
The Wilson-score interval is used, rather than a simple normal approximation, because it remains well-behaved at the small sample sizes and extreme error rates typical of compliance-attestation review samples.
Defensive Use Model
Exposure Radar is explicitly a defensive tool for the audited party. Because it applies the same extrapolation methodology a real auditor's sample-based review would use, the audited party can see its own approximate worst-case number under that methodology before an actual audit happens, and can prioritize remediation in the branches, clinicians, or payers carrying the highest projected exposure first - rather than learning its exposure for the first time from an actual audit demand letter.
Prior Art Differentiation
| Approach | Mirrors Auditor Methodology? | Conservative Counter-Figure? | Segmented Output? |
|---|---|---|---|
| Generic DRG / claims audit software (point estimate only) | Partial | No | Sometimes |
| Ad hoc internal risk scoring | No | No | Varies |
| JIL Sovereign Exposure Radar | Yes, same statistical method | Yes, Wilson-score lower bound | Yes, branch / clinician / payer |