Abstract

This paper studies whether and how an effective reminder policy can be developed in the presence of behavioural heterogeneity. We consider two types of heterogeneity: behavioural types (different individuals responding differently to a reminder) and type inconsistency (the same individual responding differently when the reminder policy changes). We take mobile data consumption as a case study. Starting from the observation that aggregate analysis can be misleading, we construct analytical models that define individual behavioural types and their optimal reminder policy using concepts of limited attention and overconfidence. We assign each mobile user to one of these behavioural types using a Bayesian mixture model and find that no single reminder policy would be optimal in our context. In a large field experiment with a corporate partner, we find that the prevalence of each behavioural type among the users is stable to changes in policy and over time. However, the behaviour of an individual user can change over time, challenging the possibility of policy personalization.