Detecting Routines in Ride-sharing : Implications for Customer Management

Routines often shape many aspects of day-to-day consumption, including transportation choice, use of mobile apps, or visits to a gym. While prior work has established the importance of habits in consumer behavior, little work has been done to understand the implications of routines, which we define as repeated behaviors with recurring temporal structures, for customer management. One possible reason for this lack of research is the difficulty of statistically modeling routines with customer-level transaction data, particularly when routines may vary substantially across customers. In this paper, we propose a new approach to measuring routine consumption, which we apply in the context of ride-sharing. We model customer-level routines with a hierarchical, Bayesian nonparametric Gaussian process, leveraging a novel kernel structure that allows for flexible yet precise estimation of routine behavior. We then nest this Gaussian process in an individual-level inhomogeneous Poisson point process, which allows us to estimate individual-level routines from transaction data, and decompose a customer’s overall usage into routine and non-routine components. We show that more routine users tend to be more valuable customers, with higher individual-level “routineness” being associated with higher future usage, lower churn rates, and more resilience to service failures