A series of recent works explore the trade-off between operation latency and data consistency in replicated data stores. In this paper, we build on the improved understandings of this trade-off and provide a new adaptive algorithm that modifies the behaviour of a quorum-based data store at runtime. In order to account for node failure we extend the existing WARS model analytically and show the accuracy of our predictions in simulations. Finally we demonstrate how our dynamic reconfiguration of the quorum sizes can be used to enhance system performance (over 30% decrease in latency) with only a small consistency cost.