AntecedentHealth
Division One · Revenue Integrity
Talk to us
Division One

Revenue Integrity

Insurance billing workflow, optimized against the treatment that was actually provided — not against the claim form somebody already typed. Findings are scoped to the date of service and every one of them cites the rule it came from.

The claim is the output. The record is the input.

Service-date scopedCoding rules change on a calendar. We apply the set that governed the date of service
Cited findingsEach flag shows the source passage from the authority it relies on
Service-line levelAnalysis at the line, not just the claim header
No API requiredWorks from an export. Most billing operations do not have an EHR API and never will
The problem, as practitioners describe it

Claims are checked in four places and still come back denied.

A claim already passes through an editor in the billing software, an edit at the clearinghouse, and an edit at the payer. Adding a fourth check in the same vocabulary at the same point in the chain does not change an outcome. We started from that objection rather than around it.

The rules are not static

Code requirements and diagnosis pairings are revised on a calendar cycle and apply to a range of service dates. Validate last year's claim against this year's rules and the finding is confidently wrong.

Every payer wrote its own interpretation

There is a nominal industry standard, and then there is what each carrier actually adjudicates against. The rules that decide whether you get paid are held by the payer, not by the code set.

The error is usually upstream

By the time a code is on a form, the decision that caused the denial has already been made — in what was documented and how the code was derived from it. Validating the form checks the symptom.

The design consequence

Field-level completeness checks belong in the billing system, at the moment of data entry, and they are not an AI problem. We deliberately do not compete for that work. What we build is the part that requires reading: whether the documentation supports the code, whether the pairing is valid for that date of service, and whether a specific payer's own published policy says otherwise.

Capability

What the system does, and what is still being built.

Stated by status, because the difference matters to anyone deciding whether to spend time on this.

CapabilityWhat it meansStatus
Procedure-to-diagnosis consistency Whether the codes on a claim support each other, at the service-line level, with pairing suggestions for codes that should be present and are not Working
Denial risk with reasons Severity-ranked findings against a retrieval corpus of authoritative coding and billing documents, each showing its source passage Working
De-identification before inference PHI is tokenized in the middle layer; the model never receives patient identity alongside the clinical detail. Access logged, MFA enforced Working
Batch processing Whole datasets processed unattended, so a backlog can be reviewed rather than a sample Working
Claim-level dialogue Ask a question about a specific claim and get an answer grounded in the same corpus Built, in testing
Patient-level, multi-claim view Analysis across a patient's claims rather than one at a time Built, in testing
Service-date rule versioning Rule sets versioned and retrieved as of the date of service, so historical claims are judged by the rules that governed them In build — priority one
Payer-specific policy Carrier-published policy as a distinct, attributable rule source separate from the general code set In build
Documentation-to-code derivation Starting from the chart note: is the code supported by what was written, and does the documentation carry the elements the payer requires? In build
Denial-history feedback Findings scored against what the payer actually did, so the system learns from real adjudication rather than from rules alone Designed, not built
Status as of August 2026. "Working" means demonstrated on real de-identified data in our environment; it does not mean deployed with a customer.
Positioning

Three things that are actually different.

Not a list of features. These are the only claims we think survive contact with someone who has spent twenty-five years inside billing operations.

01

Time is a first-class input

Most tooling answers "is this code valid?" The correct question is "was this code valid on the date this service was performed?" Rule sets are versioned and retrieved by service date. Findings on historical claims are judged by historical rules.

02

Findings carry their evidence

Every flag shows the passage from the source document behind it. That is what makes a finding appealable rather than merely assertive — and it is what lets a reviewer who is not a certified coder act on it responsibly.

03

It starts at the record

The same extraction layer that reads a claim reads the note behind it. That is the path to answering the question providers actually ask: can the visit be coded from the documentation, rather than checked after someone has coded it?

Where it fits in an existing operation

As a component, not a replacement. It sits alongside the billing platform and the clearinghouse rather than in place of either. Two insertion points are practical today: a pre-submission pass over a batch of charges, and a post-denial pass that reads the denial and the record together to identify what caused it. Neither requires an EHR integration — an export is enough, which matters because most billing operations do not have an API to the systems their clients use.

Fit

Where the economics work — and where they don't.

Where the value is real

  • Volume. The findings are incremental per claim, so the return is a function of how many claims run through them. This needs scale to matter.
  • Labor substitution. The clearest near-term value is letting someone without deep coding expertise run a reliable process, rather than adding a step for someone who already has it.
  • Recurring, known errors. A specific mistake a practice keeps making, caught systematically instead of rediscovered every month.
  • Historical review. Batch analysis over a backlog, judged by the rules in force at the time — which is only sound if service-date scoping is correct.

Where it isn't the answer

  • Low claim volume. A small practice will not recover the cost of another layer in the chain.
  • Field completeness. If the gap is required fields on a form, fix it at data entry in the billing system. That is cheaper and more reliable.
  • No access to the record. Where the chart stays in a hospital system the biller never sees, the upstream half of this cannot operate.
  • Wanting a full billing platform. This is a layer, not a system of record.

Claims we are not making

  • No published denial-rate reduction, accuracy rate, or ROI figure — we have not run a measured pilot
  • One design partner from the billing industry is engaged in review, and we're onboarding additional customers now
  • Not a certified coder, and not a replacement for one. Findings are for human review
  • Not a clearinghouse, a billing system, or a submission path
  • No guarantee of payment, appeal outcome, or regulatory compliance — the provider remains responsible for what it submits
Separation

This division does not touch the clinical side.

Revenue Integrity runs on its own corpus with its own access control. Nothing analyzed here is used for manufacturer reporting, and no clinical analysis in our other division is conditioned on reimbursement outcome. The two share an extraction layer and nothing more.