Service center automation using the state-of-the-art NLP

Description

Insurance agents send over a million emails per year, requesting endorsements, cancellations, and quotes to their in-force policies. Opening the email, understanding its intent, triaging, routing, and creating a work task for the next step of the process is very manual, which can add hours of turnaround time.

Building an email processing orchestration engine driven by text convolutional neural networks (CNNs) for classification and BERT for entity extraction yielded a significant reduction in manual processing time of inbound emails. Douglas Calegari provides an end-to-end walk-through of how his team acquired and preprocessed the right data for the training sets and evaluated and selected state-of-the-art NLP models, demonstrating how the data pipeline architecture was vital during the development lifecycle and in production.