Essays on Delay Information in Virtual Queues

Abstract

The adoption of virtual queues in the service and retail industries has accelerated greatly in recent times. In collaboration with a major ride-sharing platform, we study the wait time information (WTI), in terms of the initial magnitude of WTI and its updates as point estimates, the granularity of WTI as point estimates or interval estimates, and the design of wait time intervals, through three large-scale randomized field experiments. In the first project of Chapter 2, we study how the WTI as a point estimate, both its initial magnitude and its subsequent progress over time, impacts customers’ abandonment behavior in virtual queues. The study was conducted through a randomized field experiment that included 1,425,745 rides: one third of the rides received a neutral WTI, one-third received an optimistic WTI shorter than the neutral WTI (hence less frequent updates), and one-third received a pessimistic WTI (hence more frequent updates). The underlying wait time did not vary across the three groups. We find that both the magnitude of the initial WTI and the update frequency of the WTI have a significant impact on customer abandonment. Specifically, when adjusting the initial WTI by one minute, it did not impact customer abandonment. This is because the magnitude effect of the initial WTI is cancelled out by the opposite update-frequency effect. However, when the WTI is adjusted by more than one minute, the magnitude effect dominates: When comparing the pessimistic WTI of four minutes with the neutral initial WTI of two minutes, five minutes with three minutes, and eight minutes with five minutes, customers’ likelihood of abandonment increases by 6.2%, 14.1%, and 19.6%, respectively. Similar but opposite effects are found when comparing the optimistic WTI with the neutral WTI. We discuss how firms can use our findings and insights to design and operate better virtual queues. In the second project of Chapter 3, we study whether and how the granularity of WTI impacts customers' abandonment behavior through a randomized field experiment on our partner platform. In this experiment, we considered a point estimate, a narrow interval, and a wide interval. The point estimate was the total estimate (in minutes) based on the platform’s forecasting algorithm, whereas both the narrow and wide intervals were symmetrical and centered on the point estimate. Our preliminary results show that customers receiving the narrow interval become less likely to abandon at certain congestion levels, relative to customers receiving the point estimate. However, the wide interval does not impact customers' likelihood of abandonment. To uncover the fundamental mechanism, we propose a structural model to explore the impacts of WTI granularity on customers' prior beliefs and waiting cost-reward ratios. Our key insights show that a less granular WTI leads customers to have more optimistic prior beliefs about their wait times, as well as higher waiting cost-reward ratios. Specifically, relative to the point estimate, the narrow interval and the wide interval increase costumers’ waiting cost-reward ratio by at least 8% and 25%, respectively. In practice, we suggest that firms adopt an interval with an appropriate width to increase customers’ patience and improve their waiting experiences. In the third project of Chapter 4, we focus on the design of the wait time interval. When customers join the queue and are provided with the wait time interval, their abandonment is impacted by the magnitude of the interval, which is the center of the displayed interval, and the uncertainty about the wait time, which is measured by the width of the interval. Moreover, the service providers quote customers a promised range of their wait times, and customers might change their behaviors when their wait times exceed the lower bound or the upper bound of the wait time interval. We call this the quotation sensitivity effect. We conducted a randomized field experiment and considered the standard interval, the LB-extended interval that has the same upper bound but a smaller lower bound than the standard interval, and the UB-extended interval hat has the same lower bound but a larger upper bound than the standard interval. We find that the LB-extended interval significantly increases customers’ likelihood of abandonment by 4.19% on average over the entire time period, and the UB-extended interval does not significantly impact customers’ likelihood of abandonment on average over the entire time period. In the early stage of the waiting period, we find that both the LB-extended and UB-extended intervals increase customers’ likelihood of abandonment relative to the standard interval before customers enter any quoted intervals. This is because the increasing uncertainty effect dominates the decreasing magnitude effect in the LB-extended interval, and both the magnitude and uncertainty effects increase in the UB-extended interval. We further show the quotation sensitivity effect of the lower bound by comparing the LB-extended interval with the standard interval. We find that customers receiving LB-extended intervals become more likely to abandon when their wait times exceed their quoted lower bounds, relative to customers receiving standard intervals. A similar quotation sensitivity effect of the upper bound is found by comparing the standard interval with the UB-extended interval. We discuss how firms can use our findings and insights to design wait time intervals.