Agentic AI and RAG Enhance Straight-Through Underwriting

This research investigates the application of emerging AI architectures, specifically agentic AI and Retrieval-Augmented Generation (RAG) models, to actuarial practices, with a focus on straight-through underwriting. Actuaries are increasingly navigating a complex design space that includes traditional rule-based automation, large language models (LLMs), and sophisticated multi-agent systems capable of planning, retrieving information, using tools, and reflecting on decisions. The paper emphasizes how these advanced systems can support critical actuarial priorities such as transparency, auditability, and human-in-the-loop governance. To demonstrate these concepts, the authors developed and analyzed an agentic AI framework for the straight-through underwriting of small commercial Business Owner Policies (BOPs) within a synthetic, yet realistic, experimental environment. They compared three underwriting pipelines: a baseline single-LLM, a naive RAG system, and a multi-agent "Agentic RAG" pipeline. The agentic system, which combines targeted retrieval, third-party data checks, and explicit multi-step rule evaluation, consistently delivered the best overall performance, particularly excelling in scenarios involving multiple steps or missing information, where its structured retrieval and reflection capabilities prevented unsupported decisions.