Automatically Extracting Insurance Contract Knowledge Using NLP

Abstract

Vanbreda Risk & Benefits, a large Belgian insurance broker and risk consultant, allocates a substantial amount of time and resources to answer contract related questions from customers. This requires employees to manually search the relevant parameters in the contracts. In this paper, a solution is proposed and evaluated that automatically extracts insurance parameters from contracts using regular expressions and Natural Language Processing. While Natural Language Processing has been used in insurance for optimising premiums, detecting fraudulent claims, or underwriting, limited work has been done regarding parameter extraction. The proposed solution has been developed on 127 different contracts and two different contract types in terms of accuracy and time-performance. Moreover, the automatic parameter extraction has been compared to manual parameter extraction. We conclude that automatic parameter extraction using regular expressions achieves better accuracy than manual extraction on top of being significantly faster, allowing Vanbreda Risk & Benefits to invest more time into providing better customer service.