New CAS Research Explores LLM Applications in Claims Analysis

The Casualty Actuarial Society (CAS) has released a new research paper, Leveraging LLMs for Unstructured Claims Data Analysis, examining how artificial intelligence tools, including large language models (LLMs), may be used to extract insights from unstructured claims data such as medical records, adjuster notes, and call transcripts.

The paper presents a proof-of-concept two-stage framework designed to convert narrative claims information into structured actuarial variables. The framework separates document-level extraction from claim-level synthesis. The code converts narrative claims information into 36 structured actuarial variables with potential applications in reserving, ratemaking, and claims management. Extraction results were validated against independent clinical expert review through a “human in the loop” approach. The research explores how these approaches could expand the range of data available for actuarial analysis and support improved modeling outcomes through the application of LLM technologies.

To support transparency and further exploration, the project includes open-source code and a synthetic dataset that will be made available on GitHub.

This paper reflects the CAS commitment to equipping actuaries to lead in an AI-enabled risk landscape, where professional judgment, governance, and technical expertise remain essential to responsible use of emerging technologies.

For readers seeking broader context on artificial intelligence in actuarial practice, the CAS has also released its CAS AI Primer, which outlines key concepts, considerations, and guidance for responsible AI adoption across the insurance industry.

Leveraging LLMs for Unstructured Claims Data Analysis is available on the CAS Forum website.