Large Loss and Severity Analysis with AI

Large-loss claims are the highest-impact, hardest-to-manage files in any casualty or specialty book. A single claim can generate hundreds of pages across medical reports, legal filings, TPA correspondence, and adjuster notes. The severity drivers that determine whether a claim settles at $500K or $5M are buried across these documents, and they rarely surface in a single report.

This page covers how Parsewise helps claims and reserving teams structure large-loss files, detect severity signals early, and monitor adverse trends across an open portfolio.

The Problem: Severity Signals Buried in Fragmented Files

A large-loss claim file is not a single document. It is a document package that accumulates over months or years, typically including:

  • Medical reports, treatment plans, and independent medical examinations (IMEs)
  • Legal filings, motions, deposition transcripts, and settlement correspondence
  • TPA bordereaux, adjuster notes, and status updates
  • Expert reports (engineering, forensic accounting, vocational rehabilitation)
  • Correspondence across multiple parties and jurisdictions

Three structural problems make these files difficult to manage at scale.

Late detection of severity drivers. Reserve increases on large-loss claims often follow adverse developments that were visible in the file weeks or months earlier: a litigation status change, a new diagnosis, a shift in legal strategy. Manual review catches these signals inconsistently. By the time a file reaches the severity review queue, the reserve gap has already widened.

Inconsistent file review. Different adjusters and examiners prioritize different parts of the file. One reviewer reads the latest medical report; another focuses on legal correspondence. There is no consistent framework ensuring that every severity-relevant data point is captured, compared, and weighed across the full file history.

No portfolio-level visibility. Even when individual claims are reviewed thoroughly, aggregating severity signals across hundreds or thousands of open claims requires manual effort. Portfolio-level questions (which claims show litigation escalation? where are reserves trending adversely relative to paid losses?) depend on structured data that does not exist in freeform claim files.

How Parsewise Addresses Large-Loss Severity

Parsewise operates at the document-package level. Instead of processing individual documents in isolation, the platform ingests the full claim file and reasons across every page to produce structured, traceable outputs.

Consolidation and Structuring

Parsewise ingests the complete claim file (medical reports, legal filings, TPA bordereaux, correspondence, attachments) and consolidates them into standardized claim summaries. The platform handles PDFs, Word documents, Excel spreadsheets, scanned images, and emails. Mixed-format and multi-language documents are processed through the same pipeline, supporting the cross-border and multi-jurisdictional files common in specialty and excess lines.

Event Timeline Construction

The platform automatically builds dynamic event timelines across treatment, litigation, and claim developments. Rather than requiring a reviewer to reconstruct the chronology from scattered documents, Parsewise extracts dated events from every source in the file and assembles them into a single, navigable timeline. Each event links back to its source document, page, and paragraph.

Early Severity Flagging

Parsewise flags early severity indicators by applying cross-document reasoning across the full claim file. Severity flags are grounded in evidence from multiple documents, not inferred from a single report. Examples include:

  • A litigation status change paired with increased treatment frequency
  • Conflicting medical opinions across IME and treating physician reports
  • Reserve movements that diverge from paid loss trends
  • New legal counsel or expert retention indicating escalation

Every flag includes source citations so the reviewer can verify the underlying evidence directly.

Portfolio-Level Risk Heatmaps

At the portfolio level, Parsewise standardizes risk signals across thousands of open claims and produces portfolio-level risk heatmaps. These structured outputs allow claims leadership and actuarial teams to identify clusters of adverse development, compare severity trajectories across lines of business, and prioritize file reviews based on quantified risk indicators rather than periodic sampling.

Example Inputs and Outputs

Inputs

Document Type Examples
Medical documentation Treatment notes, surgical reports, IME reports, rehabilitation assessments
Legal documentation Complaints, motions, deposition transcripts, settlement demands, defense counsel reports
TPA data Bordereaux, adjuster status reports, payment schedules, reserve change notifications
Correspondence Claimant communications, broker notifications, multi-party emails
Multi-language files Claim documents in 70+ languages, common in cross-border casualty and specialty lines

Outputs

Output Description
Standardized claim summaries Structured profiles covering claim status, key parties, financial exposure, and severity indicators, with source citations for every data point
Dynamic event timelines Chronological sequence of treatment, litigation, and administrative events extracted from all file documents
Severity and litigation risk flags Alerts for adverse indicators (escalating treatment, litigation posture changes, reserve-to-paid divergence) with supporting evidence
Portfolio-level risk heatmaps Aggregated severity views across open claims, segmented by line of business, jurisdiction, claim age, or custom dimensions

Customer Evidence: Compre Group

Compre Group, a specialist in legacy insurance and reinsurance portfolio acquisitions, uses Parsewise to perform diligence on acquired portfolios and reconcile claims data across multiple sources. Legacy portfolios typically arrive as large, fragmented document packages spanning loss runs, bordereaux, actuarial reports, and policy documents in varying formats.

Parsewise ingests these heterogeneous document sets and automatically standardizes loss runs and reserve triangles into consistent, comparable formats. The platform reconciles paid, incurred, and reserve movements across cedants and TPAs, flagging anomalies, reserve shifts, and data gaps. Compre Group uses the structured outputs for pricing decisions, reserve adequacy assessments, and regulatory reporting on acquired portfolios.

For large-loss claims within acquired books, the platform’s ability to process the full claim file and surface severity signals at scale is particularly relevant. Legacy portfolios often contain claims with long development tails where the original severity assessment may no longer reflect current exposure. Parsewise identifies these files by reasoning across the complete document history rather than relying on summary-level data from prior owners.

How It Works Under the Hood

Parsewise’s severity analysis is powered by the Parsewise Data Engine (PDE), which processes over 25,000 pages per run with autonomous runs exceeding 5 hours. The engine breaks document layouts into subsections, contextually parses each section based on content type, and extracts entities in parallel across thousands of pages.

For large-loss files, PDE’s cross-document attention is critical. The system models relationships across the entire claim file simultaneously, catching links and contradictions that sequential, document-by-document review misses. When a legal filing references a medical event, or a reserve change contradicts the adjuster’s narrative, PDE surfaces the inconsistency with full source attribution.

Extraction agents can be configured for specific severity workflows: flagging litigation escalation patterns, tracking treatment milestones, or monitoring reserve adequacy ratios. Agents are reusable across claims and portfolios, and domain experts can create and refine them through Navi without engineering involvement.

Ready to see Parsewise in action? Request a demo or contact sales to discuss your use case.

Sources