The challenge of insurance form processing

Insurance companies often face challenges when it comes to efficiently processing complex forms that are vital to their operations. These documents, which range from applications to claims forms, contain essential business information in various formats, including structured forms, detailed tables, checkboxes and narrative questionnaires. They form the foundation for important insurance processes such as underwriting and claims processing.

Business challenges:

• High operational costs from manual processing resulting in experience delays, directly impacting customer satisfaction and customer adoption.
• Risk of human errors affecting compliance and business outcomes

Technical challenges:

• Navigate complex document layouts with mixed content types and varying quality
• Multiple form revisions and variations require flexible processing solutions
• Maintaining form-specific models and understanding contextual relationships

Recent advancements in AI and GenAI technologies offer a new path forward by automating manual form processing, reducing effort and time to market and improving document processing quality.

Introducing Intelligent Insurance Intake

HCLTech's Intelligent Insurance Intake solution, built on Amazon Web Services (AWS), is transforming how insurers handle complex forms and applications. Leveraging AWS serverless technologies to streamline document processing, combining traditional and generative AI capabilities:

Solution architecture on AWS

This serverless solution architecture on AWS features a GenAI-powered chatbot interface, where users can upload documents through the Websocket API, triggering intelligent extraction using Amazon Textract and Nova Prod backed by Bedrock agents, and saving structured results into DynamoDB. Here are the components:

1. Document storage: Amazon S3 with enterprise grade security controls ensures that sensitive insurance forms are secure and easily accessible when needed.
2. Structural AI processing: Leveraging the Amazon Textract solution, this process extracts tables, form fields and text blocks using specialized document understanding models.
3. Cognitive understanding: Using Amazon Nova Large Language Model [LLM] hosted on Amazon Bedrock, solution interpret extracted information, recognize field relationships and transform raw data into structured business information.
4. Workflow orchestration: Designed on Agentic AI architecture pattern using Amazon Bedrock Agents, solution coordinates multi-step processing through specialized agents, creating an intuitive interface where insurance professionals can interact with complex information through simple conversations.
5. Integration logic: AWS Lambda powers the workflow with serverless functions, eliminating infrastructure management and enabling automatic scaling. It seamlessly connects components and executes business logic throughout the process.
6. Data repository: To support various data formats, the solution leverages Amazon DynamoDB to store processed information in a flexible schema optimized for insurance applications.
7. Security and responsible AI: The solution enforces enterprise-grade security through end-to-end encryption (TLS 1.3 in-transit, KMS at-rest), IAM-based access controls, and comprehensive CloudTrail audit logging. HIPAA and GDPR compliance is maintained through data segmentation, automated classification and retention policies. Real-time threat monitoring via GuardDuty and security hub ensures immediate response to security events, while regular penetration testing validates control effectiveness. Solution also implements safeguards using Amazon Bedrock Guardrails to ensure Responsible AI in the application.

This hybrid approach combines deterministic document AI with contextual understanding from GenAI, delivering a level of accuracy previously unattainable with either technology alone.

Implementation highlight

1. Database schema design: Leverages Amazon Nova Pro LLM to analyze the form structure and generate a comprehensive DynamoDB schema that captures all necessary fields and relationships.
2. Form data extraction: Amazon Textract processes the document to extract raw data from tables, forms and text blocks.
3. Structured data transformation: Nova Pro's generative capabilities transform the raw Textract output into structured data that matches our database schema, understanding contextual relationships between fields.
4. Integration with Bedrock Agents: Form processing and database query agents provide a complete end-to-end experience, allowing natural language interactions with the processed data.

Advantages

The combination of Amazon Textract with Amazon Bedrock's Nova Pro LLM creates a robust solution for processing complex insurance forms with several unique advantages:

1. Zero hard-coding: No need for rigid templates or rules for each form field.
2. Contextual understanding: Leveraging Bedrock's Nova Pro LLM, solution understands relationships between different form elements.
3. Consistent database entries: A pre-defined database schema ensures consistent data processing.
4. Cross-domain versatility: This methodology can be applied to various document types across multiple industries.
5. Cost-effective efficiency: Up to 20X reduction in processing time, significantly reducing operational costs.
6. High accuracy: Solution demonstrates approximately 95% extraction accuracy across standard form fields, significantly outperforming traditional OCR and template-based approaches while requiring minimal human intervention.

Use cases:

A leading Canadian insurance provider modernizes workers' compensation form processing

Business challenges:

• Resource-intensive manual processing of forms like ACORD-130 Workers Compensation application, with each form requiring 60 minutes of staff time.
• Manual data entry error reaching 10%, significant time spent on verification, causing processing delays, impacting service quality. The high operational costs and slow turnaround times affected both customer satisfaction and business efficiency.

Technical challenges:

• Traditional Optical Character Recognition (OCR) solutions struggled with ACORD-130's mixed content types (checkboxes, tables, free text) and varying document quality, while developing AI solutions for each form type required significant time investment and expertise.
• Limited IT resources and competing priorities made it challenging to scale the solution across multiple form types while maintaining integration points.

Architecture:

• Deployed scalable AWS serverless solution starting with ACORD-130 forms, using Textract for initial form extraction and Bedrock Nova Pro LLM for intelligent data interpretation
• Implemented Bedrock Agents with configurable workflows, allowing progressive addition of new form types without major architectural changes
• Utilized DynamoDB's flexible schema and Lambda functions to create a foundation that can adapt to different form structures over time

Benefits:

• Achieved 20x efficiency improvement for ACORD-130 processing while establishing a repeatable framework for other form types
• Improved accuracy to 95% with automated validation, creating a proven model for future form automation
• Enhanced competitiveness through faster processing times and scalable architecture that supports incremental expansion to additional form types
• Drove higher Customer Satisfaction and expanded Customer Adoption

Conclusion

By implementing this GenAI-powered solution, insurance companies can dramatically reduce manual effort required for form processing, minimize errors and accelerate the underwriting process. This approach represents a significant advancement in intelligent document processing that delivers both efficiency and accuracy through the power of large language models and AWS's AI services.