EtoAI Integrity · Program-Integrity Intelligence

Pre-audit intelligence
for healthcare claims.

The first AI platform that cross-references billed claims against the clinical documentation that supports them. Every variance flag links to source claim ID, document hash, and CMS regulation citation — court-defensible by design.

VARIANCE ENGINE · LIVESAMPLE · NJ MEDICAID · PT_28A4F1
GEMINI · VERTEX AI
CLAIM SUBMITTED
CPT 99215billed
Office visit, established patient
High complexity · 40+ min
BILLED AMOUNT
$209.00
NPI 13‍72951846
DOS 2026-05-14
CLAIM_8F4C‍A219E
AI VARIANCE ANALYSIS
SOURCE NOTE“Pt presents for f/u. BP stable on lisinopril. No new complaints.Brief review of meds.Plan: continue current regimen, RTC 6mo.”
DOC-TO-BILLING MATCH0% confident
Documentation reflects a brief follow-up consistent withCPT 99213(low complexity, ~15 min). Note lacks MDM elements required for 99215.
VERDICT
VARIANCE DETECTED
− $116.00
Variance per claim · supports CPT 99213
PATTERN
47 similar encounters by this NPI in last 90d.
Est. systematic loss: $5,452
CITATION CHAIN
CMS IOM Pub. 100-04
Ch. 12 § 30.6.1
doc#a3f9b4...e2c8
Generated 1.8s · model audit ID RUN_9F2C‍13B0
READY FOR AUDIT PACKET
Google Cloud
Vertex AI · BAA active
HIPAA
Compliant · row-level isolation
FedRAMP
Path · Moderate in progress
SAM.gov
Active · UEI registered
NAICS
541512 · 541211 · 541690
THE PROBLEM

Pay-and-chase is broken.
The gap isn’t the claims data.

Every program-integrity tool on the market today analyzes claims data in isolation. None of them read the clinical documentation that legally supports those claims. The result is high false-positive rates, multi-year recovery cycles, and billions in unrecoverable losses to shell entities that have already dissolved.

$80B+
Improper Medicare + Medicaid payments / yr
GAO 2024 estimate · 8–10% of program spend
$5–10B
Recovered annually
~10× gap between detection and recovery
70–80%
Of improper payments tied to documentation issues
Not active fraud — but addressable variance
THE INSIGHT

The fraud isn’t hidden in the claims data. It’s sitting in the clinical documentation that current tools have no ability to read. EtoAI Integrity is the first platform built to close that gap at federal-scale.

THE ENGINE

Five capabilities.
One unified engine.

Built on Google Cloud Vertex AI under signed BAA. FHIR R4 ingestion for clinical data, X12 837 for claims. Citation-grounded outputs designed for federal program integrity standards.

01

Variance Engine

Cross-references every billed CPT against the unstructured clinical documentation that supports it. Vertex AI parses notes at scale; variance scores benchmarked against certified-coder ground truth.

≥90% concordancevs certified coders
02

Provider Risk Score

Continuous scoring fusing claims signal + clinical signal + network signal. Daily updates surface providers whose documentation quality drops as coding intensity climbs — well before claims-only screens trigger.

Daily refreshnot monthly
03

Program Integrity Copilot

Plain-English audit interface. Investigators type a question like “Show me providers who billed therapy while the patient was inpatient,” get cross-referenced results in seconds with every row cited to source.

Plain Englishno SQL required
04

Audit Packet Generator

One click generates a drafted Notice of Intent to Audit, evidence bundle with claim IDs and document hashes, CMS regulation citations, and a recovery estimate. Investigator reviews, signs, sends.

~8 secondsvs days manual
05

Citation Grounding

Every AI assertion is traceable to: (a) source claim ID, (b) document hash, (c) specific CMS Internet-Only Manual citation. Court-defensible by design — no black-box outputs.

100% traceableevery assertion