Transforming ISO Rating Insights with Generative AI: A Collaborative Effort

Introduction

This post is a collaborative effort by Samit Verma, Eusha Rizvi, Manmeet Singh, Troy Smith, and Corey Finley from Verisk, exploring how Verisk Rating Insights leverages generative AI to enhance user experience and operational efficiency.

The Challenge

Discussing the significant difficulties faced by users, including inefficiencies in data retrieval, manual downloads, and time-consuming support processes.

Solution Overview

An overview of the innovative solutions introduced by Verisk to overcome these challenges using generative AI technologies and AWS services, including a conversational user interface.

Key Technologies and Frameworks Used

An outline of the specific technologies employed, such as Anthropic’s Claude Sonnet 3.5 and RAG for intelligent data retrieval and response generation.

How It Works

A detailed explanation of the user experience with the new platform, focusing on its conversational interface and dynamic content retrieval capabilities.

Quality Benchmarks

A review of the evaluation framework established to ensure the accuracy, consistency, and reliability of responses generated by the system.

Business Impact and Opportunity

An examination of the significant benefits realized by both customers and Verisk, including time savings, enhanced productivity, and improved operational efficiency.

Conclusion

Highlighting Verisk’s plans for future enhancements to the Rating Insights platform and how generative AI continues to shape customer interactions and streamline processes.

About the Authors

Biographies of the authors, detailing their roles and expertise contributing to this innovative initiative.

Transforming ISO Rating Changes with AI: The Impact of Verisk Rating Insights

This post is co-written by Samit Verma, Eusha Rizvi, Manmeet Singh, Troy Smith, and Corey Finley from Verisk.

In an era where time is of the essence, Verisk’s Rating Insights feature within the ISO Electronic Rating Content (ERC) platform emerges as a powerful ally for insurance professionals. This innovative tool not only revolutionizes how users interact with ISO rating changes but also allows for efficient data retrieval in a user-friendly environment.

The Challenge: Inefficiencies in Manual Processes

Historically, accessing relevant filing information in the insurance sector meant diving into the complexities of manual downloads—this was time-consuming and, quite often, frustrating. Users faced several issues:

• Manual Downloads: To glean a single piece of information, entire packages had to be downloaded, which was both inefficient and cumbersome.
• Inefficient Data Retrieval: Identifying differences between two content packages required manual comparisons—a process that could stretch from hours to days.
• Time-Consuming Support: Verisk’s ERC Customer Support team dedicated 15% of their weekly time to assist overwhelmed customers, further complicating onboarding processes.
• Manual Analysis Time: Customers often spent 3-4 hours analyzing each test case, delaying crucial decision-making.

These challenges necessitated an innovative approach that could significantly enhance user accessibility while reducing manual workloads.

A New Era: Integrating Generative AI

To address these inefficiencies, Verisk turned to generative AI technologies using Amazon Web Services (AWS). The integration of Anthropic’s Claude AI within Amazon Bedrock has paved the way for a conversational platform that transforms how users access and analyze rating content changes.

Solution Architecture

The new architecture employs several AWS services, facilitating a smooth ingestion and inference process.

1. Data Ingestion: When a new file is uploaded, a custom chunking strategy ensures documents are properly segmented without overlap. The resulting chunks are then embedded and stored in Amazon OpenSearch as vector embeddings, allowing for efficient search and retrieval.

2. Inference Loop: This component streamlines query processing. By utilizing recent chat history, the system retrieves relevant chunks and formulates responses using the Claude AI model.

Key Technologies and Frameworks

• Anthropic’s Claude Sonnet 3.5: This advanced model provides precise insights into user queries, ensuring contextually relevant responses.
• Retrieval-Augmented Generation (RAG): This framework helps pull specific data from the embedded vector database, significantly reducing the need for extensive content downloads.

Verisk has also set up guardrails within Amazon Bedrock to maintain compliance, guaranteeing that responses adhere to quality standards.

Seamless User Experience

Now, interacting with Verisk Rating Insights is a streamlined experience:

• Conversational User Interface: Customers can pose natural language queries. For instance, "What are the changes in coverage scope between the two recent filings?" The system quickly responds with pertinent updates.
• Dynamic Content Retrieval: No more downloading entire packages; the platform facilitates immediate access to needed changes.
• Automated Difference Analysis: Queries like "Show me the differences in rating criteria between Release 1 and Release 2" yield quick comparisons without manual labor.
• Customized Insights: The intelligent output helps users grasp the impact of changes and navigate the complexities of filings effectively.

Quality Assurance: Ensuring Accuracy

To maintain high standards, a comprehensive evaluation framework is integrated into the query pipeline:

• Evaluation Framework: Validates responses for precision and relevance before delivery.
• Extensive Testing: Collaboration with subject matter experts (SMEs) ensures that outputs meet accuracy and consistency benchmarks.
• Continuous Improvement: User feedback feeds into model refinement, enhancing both accuracy and relevance.

Business Impact and Future Opportunities

The integration of generative AI into Verisk Rating Insights has resulted in remarkable outcomes:

• Significant Time Savings: Analysis time has drastically reduced, transforming what often took days into mere minutes.
• Increased Productivity: Automation allows users to focus on strategic decision-making instead of manual data retrieval.
• Reduced Customer Support Burden: The AI-powered interface empowers users to self-serve, lessening the ongoing support demand.
• Streamlined Onboarding: Efficient training reduces the need for lengthy sessions, enabling new customers to become proficient quickly.

Looking Ahead

Verisk is committed to further enhancing the Rating Insights platform. Future developments will enable more complex queries and expand functionality, capitalizing on the robustness of Amazon Bedrock to support a growing user base.

Conclusion

Through the innovative application of generative AI and AWS technologies, Verisk Rating Insights has redefined the way customers interact with and access rating content changes. By streamlining operations, enhancing accessibility, and driving efficiency, Verisk stands at the forefront of intelligent customer support and content management.

For more detailed information on this transformative journey, stay tuned for further developments from Verisk.

About the Authors

Samit Verma

Director of Software Engineering, Verisk

Samit oversees the Rating and Coverage development teams, ensuring the architectural integrity and strategic direction of multiple projects.

Eusha Rizvi

Software Development Manager, Verisk

Eusha leads technology teams within the Ratings Products division, focusing on innovative solution development.

Manmeet Singh

Software Engineering Lead, Verisk

An AWS Certified Generative AI Specialist, Manmeet drives the development of RAG-based systems, leveraging advanced AI technologies.

Troy Smith

Vice President of Rating Solutions, Verisk

Troy leads efforts behind ISO Electronic Rating Content, utilizing over 25 years of insurance technology experience.

Corey Finley

Product Manager, Verisk

With over two decades in insurance, Corey manages various products, ensuring they meet market needs.