Loss run documents contain critical claims history data needed to assess insurance risk. For Ledgebrook, extracting this data was a time-consuming, manual task prone to human error. The lack of standard formatting made it even harder to ensure consistent quality.
As a result, underwriting teams often faced delays in risk analysis and compliance reporting. GoML implemented an end-to-end AI workflow for loss run extraction pipeline, accelerating insurance automation and improving speed, accuracy, and decision-making.
The problem: slow manual processing, inconsistent data, and compliance risk
Ledgebrook’s underwriters were manually parsing loss run PDFs, each with different layouts, scanned formats, and terminology. The work was not only repetitive but also highly inconsistent. Important claims data was sometimes missed or misinterpreted, creating downstream issues in underwriting, pricing, and compliance.
The team lacked a centralized system for handling structured claims data in real time, forcing them to depend heavily on spreadsheets and manual validation. This bottleneck was a major blocker to insurance automation, slowing policy issuance and increasing error risk.
The solution: AI for insurance automation with loss run extraction
GoML built a fully automated, modular pipeline using AWS services to extract and structure claims data efficiently, a cornerstone of modern insurance automation.
Text Extraction with Textract
- Loss run documents, whether scanned or digitally formatted, were processed using AWS Textract.
- The service extracted tables, text blocks, and forms from unstructured documents, kickstarting the automation journey in insurance document workflows.
AI Segregation and Classification
- AWS Bedrock segmented the extracted content into policies and claims using a customized schema.
- This enabled clean structuring of relevant risk data and contributed to smarter insurance automation.
Data Structuring and Storage
- Each claim was broken into key-value pairs using a Bedrock-generated data dictionary.
- Structured data was stored in PostgreSQL RDS for easy analysis, auditability, and retrieval key components of automated insurance compliance.
Real-Time Access via API
- Underwriters accessed structured claims data instantly via a RESTful API (GET /loss-runs/{aiDocumentSessionToken})
- This further drove workflow automation in insurance operations.
Webhook Notifications
- Webhook integrations alerted internal systems once processing was complete.
- Streamlining operations and enhancing the broader insurance automation ecosystem.
The impact: faster insights, better risk assessment
- 70% faster processing time, reducing hours of work to minutes
- 90% increase in data accuracy, ensuring more reliable underwriting
- 50% reduction in admin overhead, freeing up time for strategic work
Lessons for other insurance and insure-tech companies
Common pitfalls to avoid
- Relying solely on PDF templates for processing
- Skipping structured data storage in favor of spreadsheets
- Delaying automation until volume becomes unmanageable
Advice for teams facing similar challenges
- Start with high-impact documents like loss runs
- Use AI to extract, structure, and store data early in the pipeline
- Ensure your API layer enables instant access for underwriters
Want to process loss run data in minutes, not hours?
Let GoML automate your loss run workflow like it did for Ledgebrook.