AI‑Driven Underwriting Workflow: A Proof of Concept using GenAI and ML

Author: Akos Arendas

Affiliation: CAS ETH – Machine Learning in Finance and Insurance

Publication date: 3 November 2025

Abstract
This paper presents a proof of concept for enhancing insurance underwriting workflows using a combination of generative AI and machine‑learning techniques. The proposed approach integrates a retrieval‑augmented generation (RAG) system for guideline‑based knowledge retrieval with a similarity‑assessment framework designed to identify comparable historical insurance cases. The RAG component leverages large language models and vector search to enable fast, context‑aware access to internal underwriting documentation, while the similarity engine applies dimensionality‑reduction and clustering methods to support consistent technical decision‑making. Applied within a global reinsurance underwriting context, the solution demonstrates how modern AI techniques can reduce manual document review, improve standardisation, and free underwriters to focus on higher‑value analytical tasks. The paper discusses system architecture, implementation choices, and limitations, and outlines directions for further validation and production‑grade deployment in regulated environments.