A multi-input multi-label claims channeling system using insurance-based language models. (15th September 2022)

• Record Type: Journal Article
• Authors: Dimri, Anuj; Paul, Arindam; Girish, Deeptha; Lee, Peng; Afra, Sardar; Jakubowski, Andrew
• Is Part Of: Expert systems with applications. Volume 202(2022)
• Publication Date: 2022-09-15
• Subjects: Insurance -- Language models -- BERT -- Transfer learning -- Claims classification -- Fraud detection; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.117166
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Servicing claims, a time consuming and labor-intensive task, plays a pivotal role in how insurance companies serve their policyholders. Claims may not get routed early enough in the process to the correct team, leading to dissatisfied customers because of inefficient claim's management. Claims departments need to process substantial amount of structured and unstructured data to successfully route claims — a process referred to as channeling. The scope of the present work is limited to the auto insurance claims with a focus on four different downstream classification tasks including claims' fraud and bodily injuries. We propose a system that utilizes claims' notes and structured data to build machine learning models, which employ an insurance-based language model built by enhancing Google's BERT, to route claims to domain experts. The proposed channeling system successfully routes important claims to domain experts for additional review, which can substantially improve claims management and customer satisfaction. Highlights: A channeling system to improve claims management and customer satisfaction. Built models using structured data and claim notes for claims classification. Compared multi-label approach with an equivalent binary approach for multiple tasks. A web-based user interface to serve the model(s) and route claims to domain experts.