We present a general Bayesian formalism for the definition of figures of merit (FoMs) quantifying the scientific return of a future experiment. We introduce two new FoMs for future experiments based on their model selection capabilities, called the decisiveness of the experiment and the expected strength of evidence. We illustrate these by considering dark energy probes and compare the relative merits of stages II, III and IV dark energy probes. We find that probes based on supernovae and on weak lensing perform rather better on model selection tasks than is indicated by their Fisher matrix FoM as defined by the Dark Energy Task Force. We argue that our ability to optimize future experiments for dark energy model selection goals is limited by our current uncertainty over the models and their parameters, which is ignored in the usual Fisher matrix forecasts. Our approach gives a more realistic assessment of the capabilities of future probes and can be applied in a variety of situations.