Text Analysis with BERT : Automatic Identification and Classification of Insurance Offer Rejections

Text analysis with BERT: Automatic identification and classification of insurance offer rejections. E-mails have become the prevalent way of communicating between companies, their partners and their customers across many industries. In insurance brokerage, incoming e-mails containing quote or cost proposal rejections are currently processed and categorized manually, requiring significant time and resources. This study aims to investigate whether machine learning-based language models can be used to accurately automate the identification of rejection causes. The study examines the potential for reducing time and resource consumption in the workflow of a company by introducing a BERT model trained for multi-class categorization. Training data consists of 3400 labeled e-mails, originally unlabeled and unformatted. The data was classified for training usage using GPT-api and later manually adjusted. The dataset was split 80/20 for training and testing, and model performance was evaluated using accuracy, precision, recall and F1-score. The model achieved an accuracy of 0.934, precision of 0.937, recall of 0.934 and an F1-score of 0.935. The study finds that BERT performs very well on the given text classification task, and that there is potential for increased productivity and time savings in the workflow when implementing the model.