A Predictive Investigation of First-Time Customer Retention in Online Reservation Services

Abstract

This paper reports a predictive investigation of first-time customer retention in an emerging service business—online reservation services. We work with an online platform that enables customers to make reservations for various types of restaurants. With numerous first-time users on the platform, the focal company is eager to effectively identify recurring customers. However, the business problem is challenging due to that each first-time customer has one and only one booking record hinders the use of well-established marketing models that demand multiple booking records for a customer. By analyzing more than 100,000 observations, we extract booking-related features that are useful in predicting first-time customer retention. Our feature extraction is potentially applicable to other service sectors (e.g., hotel, travel) with similar booking information fields (e.g., reservation timing, party size). We further conduct a comparative study in which surprisingly, the seemingly simplistic generalized additive model (GAM) for our test cases consistently outperforms computationally intensive ensemble learning methods, even the cutting-edge XGBoost. Our analysis indicates that there is no silver bullet for applied predictive modeling and GAM should definitely be included in the arsenal of business researchers. We conclude by discussing the implications of our study for online service providers and business data analytics.