Queueing to Learn∗

I study the design of a queue when agents learn from past service experiences and can be served multiple times. A continuum of forward-looking agents compete for a constant flow of a resource and decide whether and when to engage in costly queueing to be served. Valuations fluctuate over time, independently across agents; each agent faces an experimentation problem because payoffs are informative about the prevailing valuation. To maximize efficiency, the designer offers a simple binary menu of queues (i.e., two customer classes): service is rendered on a first-come first-served basis in one queue, and in random order in the other. Surprisingly, allowing for strategic reneging, the designer can implement the optimal menu with a single queueing discipline.