Bayesian implementations of mark-recapture and occupancy models are become increasingly common in ecological studies. Models are frequently implemented using intuitive code that allows users to modify and extend models. Specifically, models are coded using discrete latent states. While flexible, these approaches to population and movement models can be exceedingly slow, especially when data sets are large and sparse. Slow models can lead authors to either forego or limit the size of simulation studies associated with new models (potentially leading to the adoption of methods that have not been well tested) or to only consider an artificially small subset of potential models. In some instances, users may simple walk away from a Bayesian approach if the model fails to compile reasonably quickly. Here we show how to marginalize out discrete latent states and the consequences for the time required to fit models. We explore a number of case studies illustrating both the relative quickness of this newer approach, but also some of its limitations when considering more complex, and ecologically feasible, models.