AI Loss Run Parsing Agent

How Cassidy automates loss run parsing using AI

Step 1: Trigger on new loss run submission

The Workflow activates when a loss run arrives—whether uploaded by a broker, received via email, or pulled from a carrier portal—and classifies it by account, carrier, and line of business.

Step 2: Extract and structure claim data

Cassidy processes the document using OCR and table detection, capturing claim-level fields including dates, paid amounts, reserves, cause of loss, status, and jurisdiction—even from scanned PDFs with stamps or watermarks.

Step 3: Normalize to your underwriting schema

The Workflow maps carrier-specific terminology to your standard data model, harmonizing labels like “total incurred” vs. “claims paid,” splitting AL/APD and GL sub-coverages, and consolidating multi-line entries per claim.

Step 4: Validate with Human-in-the-Loop review

Confidence scoring flags uncertain fields for underwriter attention. Validation rules catch logical errors—like reserves exceeding incurred—and route exceptions to a review queue with full audit trails.

Step 5: Generate frequency/severity analytics

Cassidy aggregates claims by LOB, policy year, cause, and location to surface KPIs: claim frequency, average severity, open reserves, large loss counts, and trend comparisons—delivered as structured data or dashboard-ready outputs.

Step 6: Push insights to your underwriting workbench

Clean datasets feed directly into your rating models and underwriting systems. Narrative summaries highlight trends, hotspots, and recommended risk controls—ready for the underwriter’s decision.