The inconsistencies are already in the record. Surfaced, cited — your SIU makes the call.
Medical claims fraud detection software from Medrecords AI flags contradictions within a claim and patterns across claims — cloned provider notes, upcoding signals, mechanism-of-injury drift — each with citations, and routes them to human review. Every flag ships as a signal, never an accusation: your SIU investigator decides what it means.
Signals, never accusations: every flag ships as a cited, referral-ready SIU packet, and a human investigator decides what it means. Mandatory human review before any action — no automated determinations, ever.
Contradictions within the claim, lined up and cited.
The record often disagrees with itself: the intake describes one mechanism of injury, a later note describes another; a history denies what an earlier page documents. Because the platform reads every page, these collisions surface automatically — each one presented as two cited passages, side by side.
Patterns a single file can't show you.
One claim looks fine on its own. Across the portfolio, the same paragraph in six different claimants' notes, or a billing level that never varies with the visit, becomes visible. Cross-claim signals compare provider documentation across your whole book and cite the specific documents where the pattern repeats.
A referral packet your SIU can verify line by line.
When a signal is worth escalating, it leaves the platform as a referral-ready packet: the flagged inconsistency, the source pages behind it, compared passages side by side, and the claim identifiers involved. No score and no conclusion — the investigator's judgment is the product; the packet is its raw material.
An accusation is a human act. This tool doesn't make them.
A fraud referral affects a provider's livelihood and a claimant's benefits, which is why every signal here is a cited, checkable fact about the record — not a risk score. Your SIU can verify each one against the source in seconds, and nothing moves on a claim without a person deciding it should.
See Record Version & Alteration DetectionFrom claim file to SIU-ready signal.
Three steps, with humans owning the one that matters.
Records, bills, and correspondence are read page by page — a single claim or the whole portfolio.
Intra-record contradictions and cross-claim patterns are flagged, each with the exact pages behind it.
Your team clears the signal or escalates it as a cited SIU packet. Every outcome is a person's call, on the record.
Who works signals with it.
Teams that carry the duty to investigate — and the duty to get it right.
SIU teams get cited, checkable referrals instead of hunches — and adjusters get contradictions surfaced before settlement, not after.
For carriersConsistent signal screening across every client's book, with referral packets each carrier's SIU can act on in its own process.
For TPAsFraud signal detection, answered.
No — and it never will. The platform surfaces signals: documented inconsistencies and patterns, each cited to the exact pages involved. Whether a signal means sloppy templating, a data-entry error, or something worse is a determination only a human investigator can make, and mandatory human review sits before any action. No score, no verdict, no automated decision.
Yes. Beyond a single claim, it compares provider documentation across your claim portfolio — near-identical narrative text appearing across different claimants, recurring billing-code patterns, and repeated report structures. Each cross-claim signal cites the specific documents in the specific claims where the pattern appears, so an investigator can pull them side by side.
Every flag ships referral-ready: the signal, the exact source pages it rests on, the passages side by side where text is compared, and the claim identifiers involved — assembled as a cited packet your SIU can verify line by line. What it deliberately does not contain is a conclusion; the investigator writes that.
A cloned-note signal fires when narrative text in one visit note is near-identical to another — a different date, a different patient, the same paragraph. An upcoding signal fires when the billed service level and the documented encounter don't line up, with both the billing line and the clinical note cited. Both are common, both have innocent explanations sometimes — which is exactly why they're flagged for review rather than judged.
The stated cause of injury changing shape as the claim ages — the intake note describes one event, a later provider records a different one, and the demand letter tells a third version. The platform lines the versions up chronologically with each one cited to its page, so the drift is visible as documented fact rather than an impression.
Related capabilities.
Signal detection reads the same cited record the rest of the platform builds.
Every claim in an incoming demand checked against the record.
ExploreBilled charges benchmarked against UCR data, outliers flagged with context.
ExploreWhen near-duplicate pages differ, the delta is the finding.
ExplorePortfolios screened against your criteria; judgment calls surfaced to humans.
ExploreSee what the record has been trying to tell you.
Upload a claim file and get back its documented inconsistencies — cited, checkable, and ready for your reviewers to judge. Handled under our BAA; never used to train a model.