Back

Ledgebrook Automating Liability Code Extraction from ACORD Documents

Deveshi Dabbawala

February 24, 2025
Table of Content

Business Problem

  • Time-Consuming Data Entry: Manually extracting liability codes from ACORD forms was labor-intensive and error-prone. 
  • Inconsistent Business Classification: Errors in classification led to incorrect risk assessments, impacting underwriting decisions. 
  • Slow Underwriting Process: The lack of automation caused delays in policy approvals and premium calculations. 
  • Scalability Concerns: As document volume increased, manual extraction became unsustainable for growth.  

About Ledgebrook

Underwriters rely on NAICS, SIC, and Class Codes to assess business risk. Extracting and categorizing this information manually from ACORD forms (125, 823, 126) was inefficient. goML developed an AI-powered liability code extraction service. 

Solution

goML designed a high-speed liability code extraction system using AI-powered automation: 

Automated Text & Data Extraction: AWS Textract processed ACORD forms to extract business classification details. 

Centralized Storage & Searchability: Processed document details were stored in PostgreSQL RDS, and documents were vectorized and stored in OpenSearch for quick lookups. 

Code Matching & Categorization: Extracted data was mapped to standardized NAICS, SIC, and Class Codes, ensuring accuracy. 

Search & Compliance Reporting: OpenSearch Serverless provided real-time search capabilities, improving compliance and reducing misclassification risks.

Lambda-Powered Processing: AWS Lambda functions processed and validated the extracted classification data. 

Architecture

  • User Interaction & API Triggers 
    POST /store-liability-codes → Triggered inside the Document Service
    User sends a request with an aiDocumentSessionToken
  • Retrieval & Processing of Liability Codes 
    Bedrock & OpenSearch build a Retrieval-Augmented Generation (RAG) Pipeline to: 
    Extract addresses and business types from non-loss run document chunks
    Retrieve liability codes from an Excel sheet stored in AWS S3
  • Data Structuring & Storage 
    Extracted liability codes are stored in the database (PostgreSQL or similar DB)
  • Webhook Integration & Execution 
    Triggers a Webhook Endpoint after processing is completed. 
    Passes the Liability Code Response to the Document Service for execution. 
  • Data Retrieval & Response Generation 
    GET /liability-codes/{aiDocumentSessionToken} allows users to fetch the liability codes. 
    Query retrieves the stored Liability Code Response from the database. 
    Returns structured response to the user. 

Outcomes

80%
Faster underwriting decisions
95%
Accuracy in business classification
60%
Improvement in operational efficiency