On Accommodating Customer Flexibility in Service Systems

We consider simple parallel queueing models in which a proportion of arriving customers are flexible, i.e. they are willing to receive service at any one of some subset of the parallel servers. For the case of two parallel servers, we show that as the servers become fully utilized, the maximum improvement in mean waiting times is achieved for arbitrarily small levels of flexibility. The insights from this analytic model are supported by simulation results that show that large gains can be made with low levels of flexibility. The potential implications of these results for two motivating examples are discussed.