Empirical Analysis of Referrals in Ride-Sharing

Firms often offer the option to refer friends in exchange for a reward. In this paper, we empirically address the question of how service usage---in terms of experience level, current usage intensity, and recency---affects the probability of making referrals and the quality of those referrals. We incorporate dynamic behavior in our models to analyze how past referrals affect future referrals. We partner with a ride-sharing platform, allowing us to access a large panel dataset on transactions and referral actions. We estimate econometric models that account for unobserved heterogeneity to show that the probability of making a referral increases with the experience level (captured by the number of past rides), increases with the current usage intensity (number of rides in the previous week), decreases with long inactivity periods, and decreases with past high quality referrals. We also find that referral quality---measured by the number of rides completed by the referred customer---increases with experience and decreases with past high quality referrals. Finally, we consider a prescriptive campaign in which the platform sent notifications to remind users about the referral program. Using data from a field experiment, we show that such notifications can increase referral rates by 46% and generate significant marginal revenue.