Select a fraud analytics vendor (e.g., Verisk CAIVRS, Shift Technology, or a carrier-specific model); obtain API credentials and the scoring endpoint URL from the vendor's integration guide.
At FNOL or claim update, assemble the fraud-scoring payload: claim type, loss date, loss description, claimant demographics, policy history, prior claims count, and any telematics or IoT signals available.
POST the payload to the vendor's scoring endpoint; receive a fraud score (numeric or tiered: low/medium/high) and a list of contributing indicators in the response.
Apply your carrier's referral threshold (consult your SIU guidelines for the exact score cutoff) to determine whether to auto-refer to the Special Investigations Unit or continue normal processing.
Write the fraud score, indicator list, and referral decision back to the claim record as structured metadata; trigger an SIU workflow task if the threshold is met.
Log all scoring calls with input hash, score, model version, and timestamp to satisfy regulatory audit requirements and support adverse-action documentation.
Known gotchas
Fraud scores based on demographic or protected-class data can trigger fair-claims-handling violations; validate vendor scoring features against your state's unfair discrimination statutes before deployment.
Fraud model accuracy degrades over time as fraud patterns evolve; schedule periodic model re-evaluation with your vendor and track false-positive and false-negative rates against closed SIU outcomes.
Auto-denying claims based solely on a fraud score without human review is legally risky in most jurisdictions; the score should route to SIU for investigation, not serve as a standalone denial basis.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp