Payment Integrity / claims audited against the chart

Most audits read the claim.
We read the chart it came from.

Roughly a quarter of U.S. health spending — an estimated $760 to $935 billion a year — is lost to waste, fraud, abuse and error. The tools built to catch it see only the bill, so they infer problems from billing patterns. CareCompile ingests the billing feed on the same wire as the orders, results and notes that justify it. So every charge gets the one check pattern analytics can't run: is the clinical evidence actually there?

The audit runs in both directions charges without evidence/duplicates, unbundling, upcoding, phantom billing evidence without charges/documented care that never produced a bill denials worth fighting/appeals drafted from chart evidence every verdict human/the system assembles; your team decides
Start a silent audit → See what it catches
Charge audit · review queue SYNTHETIC VALIDATION STREAM
WhenWhat the engine sawState
09:41
80053 CMP, billed three times
same panel · same visit · same service date
Duplicate
09:44
99285 ED visit, level 5
documentation on the wire supports level 3
Upcoding
09:52
71046 chest X-ray, two views
no radiology report anywhere in the record
No evidence
10:15
80048 BMP beside 80053 CMP
the basic panel is a subset of the comprehensive one
Unbundled
10:22
Central line placed, per note
no corresponding charge within the capture window
Missed charge
10:38
Case file assembled
charge · clinical artifacts · drafted audit memo
For review
10:51
Denial CO-50, medical necessity
appeal drafted from the evidence already on the wire
Appeal ready
11:04
Reviewer verdict recorded
flagged · by name · tamper-evident audit trail
Human call
~1/4
Of U.S. health spend
is waste (JAMA)
Both
Directions
of leakage
30
Days beside your
feed, silently
0
Real patient records
used to build it
Synthetic validation stream shown. Findings are leads for human review — never automated determinations.
01 — THE GAP

The evidence isn't in the claim.
It's in the chart.

Whether a charge is legitimate is a clinical question, and claims data can't answer it alone.

Two ways to audit the same charge
The questionClaims-side analyticsAudited against the chart
Did the service happen? Inferred — a provider whose imaging volume is a statistical outlier gets a second look, eventually. Checked — an imaging charge with no report, no order and no mention in the record is flagged on that visit, individually.
Was it coded at the right level? Inferred — an E/M distribution skewed toward level 5 across months of claims suggests upcoding. Checked — the documentation that arrived on the same wire either supports the billed level or it doesn't.
Was it billed once? Caught when the duplicate crosses claims — missed when it hides inside one. Caught at the charge line: same code, same visit, same service date, no distinct clinical event behind the repeat.
What never got billed at all? Invisible. A charge that was never submitted leaves no claims trail to analyze. Visible — the procedure note is on the wire. If no charge follows it inside the capture window, that gap is surfaced.
When do you find out? Post-payment, then pay-and-chase. Pre-bill by default — the audit runs while the charge is still in the building. Post-payment review uses the same engine.
Why we can
do this

CareCompile was a clinical engine before it was an auditing engine. The platform already parses admissions, orders, results, documents and billing transactions from the same HL7 v2.x and FHIR R4 stream, and fuses them into one patient record. Payment integrity is that same fused record, read with a different question: not "is this patient deteriorating?" but "does the chart support this bill?" The interface work that takes claims vendors months — getting clinical context flowing next to billing data — is where this engine started.

02 — WHAT WE CATCH

Three kinds of leak.
One engine.

Improper charges leaving, earned revenue never billed, and payable claims denied — all three are chart-versus-bill questions.

01
Charges the chart can't support
Fraud, waste & abuse detection

Every charge line is checked against the clinical record it arrived with. The classic patterns — and the one that matters most: services billed with no clinical evidence they happened.

  • Duplicate charges — same code, visit and service date
  • Unbundling — component codes billed beside their panel
  • Upcoding — billed level vs. documented complexity
  • Phantom billing — no order, result or note behind the charge
  • Impossible days — units and time that exceed the calendar
  • Price variance — charged vs. expected allowed amount
02
Care the bill never captured
Charge-capture recovery

The same cross-check, reversed. A procedure note, a positive result, an administered dose — clinical events that should generate a charge and didn't, surfaced within configurable time windows by department and event type.

This is the half claims-side tools cannot see at all: a charge that was never submitted leaves no claim to analyze. For most provider organizations, the recovered charges are the business case for the entire engagement.

03
Denials the chart can win
Evidence-based appeals

When a claim comes back denied for medical necessity or coding, the evidence for the appeal usually exists — scattered across the record. The engine assembles it: the order, the result, the documentation, the timeline — and drafts the appeal letter from what is actually in the chart, for your billing team to review and send.

Denials that the chart genuinely can't support are worth knowing about too — they point back to the documentation gap that caused them.

03 — THE CASE FILE

A flag is not a finding.
A case file is.

Anomaly scores don't survive an investigation. Evidence does — so evidence is what the system hands your team.

STEP 1
Detect

Deterministic rules run on every charge as it arrives — explainable checks, not a black-box score. Each flag states its reason in plain language: which codes collided, which documentation is missing, which window closed unbilled.

STEP 2
Assemble

For each flag, the AI copilot gathers the complete case: the charge line, every clinical artifact that supports or undermines it, the patient timeline, and a drafted audit memo a reviewer can accept, amend or reject.

STEP 3
Decide — a human does

Your reviewer flags or clears each finding by name, in a queue built for the workflow. Every verdict lands in a tamper-evident audit trail — who reviewed what, when, and on what evidence. The system never decides anything alone.

Findings are leads, not accusations — whether something is an error, a documentation gap or a matter for investigation is always your team's call.

What this deliberately is not

This is not an automated fraud accuser, and it is not a denial engine. No charge is blocked, no claim is auto-denied, no provider is scored in secret. The system's whole job is to put the right evidence in front of the right person early enough to matter — the judgment stays where it belongs, with your coding, billing and investigation professionals.

04 — PROOF

We manufacture the fraud.
So we know the answer key.

A detection claim you can't score against ground truth is a marketing claim. Ours is scored before your data is ever involved.

The validation
loop

MediFlow, our clinical simulation engine, generates the audit's exam — and its answer key. It produces admissions, deteriorations, lab results, documentation and billing transactions as real HL7 and FHIR traffic, including deliberately seeded integrity defects: duplicated panels, unbundled codes, upcoded visits, charges with no supporting documentation, documented procedures with no charge. Because the defects are planted, detection performance is measured — what was caught, what was missed — not asserted.

It is also why no real patient data has ever touched development or validation. The engine arrives at your door already exercised against thousands of synthetic charge lines, and the silent audit period then calibrates it against the realities of your own feed — your charge master, your documentation habits, your payer mix.

Seeded defects, planted then found
Measured · duplicates, unbundling, upcoding, phantom charges
Charge-capture gaps
Measured · documented events with no bill inside the window
Your feed, silently
Calibrated · 30 days shadowing production before any readout
05 — ENGAGEMENTS

Three ways in.
All three start silent.

Every engagement begins the same way: the engine runs beside your feed for thirty days, and then we show you what it found.

FOR PROVIDERS · 01
Pre-bill charge audit
Catch it before it leaves the building

Duplicates, unbundled panels, unsupported levels and undocumented charges caught while the claim can still be fixed — before it becomes a denial, a takeback or an audit finding with your NPI on it.

Starts with30-day silent audit beside the live billing feed
You getA readout: what it would have flagged, and what it stayed quiet about
ThenThe review queue goes live inside your billing workflow
FOR PROVIDERS · 02
Charge-capture recovery
The leak that runs in your favor to fix

Documented care that never produced a charge, surfaced while it is still billable. The recovered charges are the business case — this is the engagement that funds the others.

Starts withCapture-gap rules tuned to your departments and event types
You getA queue of documented, unbilled events with the evidence attached
ThenOngoing monitoring so the gap stays closed
FOR PAYERS · 03
Payment integrity review
For plans, TPAs and self-insured employers

Post-payment FWA detection over claims with clinical context where you have it — every finding delivered as an evidence-backed case file ready for special-investigation review, not a spreadsheet of scores.

Starts withA retrospective pass over a claims window you choose
You getRanked case files: evidence, reasoning, drafted memo
ThenContinuous review as claims arrive
A working session, not a pitch

Run it beside your billing feed for thirty days.

No workflow change, no charge blocked, nobody's process interrupted — the engine just watches the same wire your bills already travel. Then we sit down with your revenue-cycle or investigation team and go through the readout together: what it flagged, the evidence behind each flag, and what it stayed quiet about. If the findings don't justify the engagement, that will be obvious to both of us.

Start the conversation → Ask an engineering question

Payment Integrity Notice — CareCompile's payment-integrity capabilities are decision-support tools for qualified billing, coding, compliance and investigation professionals. All findings are advisory leads for human review; they are not determinations of fraud, abuse or improper payment, and no finding results in an automated denial, charge block or adverse action. Coding and billing decisions remain the sole responsibility of certified coding and billing staff, and investigation decisions remain with the responsible compliance or special-investigation personnel. CareCompile does not provide legal advice. Waste estimate: Shrank, Rogstad & Parekh, "Waste in the US Health Care System," JAMA, 2019.