Leveraging unstructured call log data for customer churn prediction

Customer retention is important in the financial services industry. Machine learning has been incorporated into customer data analytics to predict client churn risks. Despite its success, existing approaches primarily use only structured data, e.g., demographics and account history. Data mining with unstructured data, e.g., customer interaction, can reveal more insights, which has not been adequately leveraged. In this research, we propose a customer churn prediction model utilizing the unstructured data, which is the spoken contents in phone communication. We collected a large-scale call center dataset with two million calls from more than two hundred thousand customers and conducted extensive experiments. The results show that our model can accurately predict the client churn risks and generate meaningful insights using interpretable machine learning with personality traits and customer segments. We discuss how these insights can help managers develop retention strategies customized for different customer segments. � 2020 Elsevier B.V.