The platform

Watch one claim move through Grivara.

Four acts. Seven scenes. Ten specialists. One decision packet. Scroll to see how an ambiguous health claim at Northvale goes from PDF to signed audit trail — in minutes, not weeks.

Scroll to begin
TenantNorthvale Health
ClaimCLM-MED-0010
Line of businessHealth · Outpatient
Exposure$4,800
StateIn pipeline

Scene 1 · FNOL arrives

A claim lands as documents, not data.

9:14 AM. A new claim hits Northvale's intake queue — one EOB, one itemized bill, one referral. No structured fields yet. Everything is prose and PDF.

Inbound envelope
3 documents · received via portal · PHI removed before review
Portal upload9:14 AM · Apr 12
CLM-MED-0010.pdf
eob.pdfitemized-bill.pdfreferral.pdf

EOB · Northvale

Patient: REDACTED (member #••••4821)
Dx: H66.9 — Otitis media, unspecified
Procedures:
  70551  MRI brain, without contrast
  76536  Ultrasound, soft tissue neck
Billed: $4,800.00

Scene 2 · Extraction

Documents become a graph, not a table.

Extraction doesn't just pull fields. It connects every piece of the claim — member, provider, diagnosis, procedures, referring doctor — into a live map of who touched what. The moment the provider's ID lands, the map already shows two prior flagged claims from the same provider. Fraud rings surface the instant they form, not weeks later when a pattern finally becomes obvious.

Live claim map
6 connected parties · 1 flagged match
NPI 1528H66.9Mem 4821ReferrerCPT 70551CPT 76536CL·0007FLAGGEDCL·0010

Provider 1528 is already linked to a prior flagged claim (CL·0007). The ring is visible before the adjuster opens the file.

6 entities · 7 edges·2-hop flagged

Scene 3 · Specialists fire

Four specialists. Four findings. Every one cited.

Coverage, Evidence, Fraud, and Compliance run concurrently. Each one cites the policy clause, registry, or guideline it relied on. Your adjusters don't get a verdict — they get a finding with its sources next to it. No guesswork, no black box.

Specialist findings
Every finding is cited · every citation is auditable
Claim CLM-MED-0010 · specialist review
4/4
All specialists safely preparedPHI removedInput verifiedApproved model
Coverage Analystin review

Evidence Collectorin review

Fraud Signal Analystin review

Compliance Agentin review

Reviewed 9:17 AM · ready for governance

Scene 4 · Runtime gates

Nothing runs without passing the gates.

Seven approval gates check the specialists' work against Northvale's own rules. Required evidence is present. Medical necessity fails on imaging. The fraud signal lands above the review line. One gate flips.

Approval checks
Configured by Northvale · fully auditable
PII scrub
PendingSafety
Input quality
PendingSafety
Required evidence
PendingApproval rule
Medical necessity
PendingApproval rule
Min confidence
PendingApproval rule
Approved models
PendingApproval rule
Human review
PendingTenant rule

Scene 5 · HITL routing

Human at the keyboard, not in the loop forever.

Because one gate flipped to review, the claim routes to a medical director. Not auto-denied. Not auto-approved. The specialist sees the full agent trace, the cited policy text, and one-click options: approve, request records, deny with reason, or escalate to SIU.

Specialist Inbox
1 open · target 3 days · current average 18
New review · CLM-MED-0010

Medical Director · Inbox

CLM-MED-0010 · MRI + US for otitis media

Elevated fraud · medical necessity fail

Approve with conditions
Request records
Deny
Escalate SIU

Scene 6 · Decision packet

The packet is the product.

Every claim ends with the same record: what each specialist found, the policies and guidelines they cited, the coverage call, the recommended action, and a complete history your team can replay. This is what regulators see. This is what auditors sign.

Decision record
Replayable · signed · version 12

Decision record · CLM-MED-0010

Claim
CLM-MED-0010
Coverage
Question — prior auth missing
Fraud signal
Elevated
Flags
Dx mismatch · Imaging · Upcoding
Cited
Policy §4.2 · Clinical guidelines
Recommended action
Request more records
Review required
Medical director
Audit
Replayable · signed
Signed · v12

Scene 7 · Action + audit

Closed loop. Signed trail.

Once the medical director approves, the action fires into your core system: request additional records, hold billing, route to medical review. Every step — what each specialist did, what the gates decided, who approved — lands in a signed, replayable record.

Action to core system
Sent · delivered · receipt captured
POST/claims/CLM-MED-0010/actions
{  "action": "request_documentation",  "required": ["medical_records", "referral"],  "hold_billing": true,  "signed_by": "md.carter",  "config_version": "v12"}
Sending…

A real pilot case from the Northvale synthetic batch. Ground truth: pend for medical review, fraud signal present, 10-day cycle target.

The roster

Ten specialists, one decision packet.

No generalist model. Each agent owns a narrow job, produces a structured output, and hands off to the next.

01

Intake Router

Triages line of business, acuity, and priority.

02

Extraction Agent

Turns documents into structured, cited facts.

03

Coverage Analyst

Applies the active policy version, cites clauses.

04

Evidence Collector

Pulls provider NPI, weather, history, CRM.

05

Fraud Signal Analyst

Catches organized fraud rings, not just lone red flags.

06

Compliance Agent

HIPAA, state rules, timing, appeal rights.

07

Customer Comms Agent

Plain-language notices, EOBs, escalations.

08

Reserve Analyst

Sets reserves against authority thresholds.

09

Orchestrator

Synthesizes specialists into one recommendation.

10

Action Executor

Sends decisions to your core system and captures the receipts.

Surfaces

Four ways the product meets your team.

One backend. Four front doors, each tuned to a role.

Claims Cockpit

Your adjuster's single screen. Timeline, citations, evidence, approve — all in one place.

Specialist Inbox

Every claim that needs a human looks the same: reason at the top, one-click actions, no hunting for context.

Policy Console

Upload, version, activate, rollback. Chat against any version with citations.

Audit Trail

Signed, replayable record of every decision, override, and approval.

Your data, your rules

Your book stays yours. Every decision replayable.

Your claim data is isolated from day one. Pick the AI vendors your security team has already approved. Set your own thresholds. Bring your own data. And replay any decision an auditor asks about.

01

Your book stays yours

Your claim data lives in its own workspace — not a shared pool. No other carrier, TPA, or health plan on Grivara can see a byte of it. Isolation is built in, not layered on with permissions.

02

Use the AI you trust

Run Grivara on OpenAI, Anthropic, or a private deployment — whichever your security team has already cleared. Swap vendors whenever you want. Your specialists don't need to be retrained.

03

Replay any decision

Every decision is sealed with the exact rules and data it used. Pull up a claim from last quarter, run it again with today's policy, and see what changed. Show an auditor the chain of reasoning from six months ago.

04

Your data, protected

Claim files stay encrypted. Patient and personal information is removed before any AI model sees a word of it. Your audit log can be read — but never edited.

Where it fits

Grivara vs. the alternatives.

CapabilityLegacy coreBPOGeneric copilotGrivara
End-to-end claim pipeline
Cited decisions, no black box
Pre-flight + real-time approval gates
Human approval on deny / SIU / over-authority
Signed, replayable audit trail
Specialists tuned per line of business
Actions that land in your core system
Minutes to decision, not weeks

Walkthrough

See your next claim move in real time.

Thirty minutes. We bring a live pipeline wired to a synthetic tenant like Northvale. You bring one claim — or let us run CLM-MED-0010 again.

  • Live decision record
  • Approval rules in action
  • Human review preview
  • Replay the audit trail
Back to homepage

No install. No PHI in the demo. NDA on request.