Predictive, data-driven models are of central importance for the application of artificial intelligence in sport. The data situation in sport is characterized by small numbers of cases, sometimes high noise and incomplete data, which make the use of suitable machine learning methods necessary. Bayesian inference is considered a suitable approach to deal with this problem. With normally distributed parameters and data, exact inference is possible with Gaussian processes, for example. Marcov chain Monte Carlo methods allow the use of arbitrary distributions for prior and likelihood. By using suitable priors and selecting models that are not unnecessarily complex, meaningful predictions can be made even with small and incomplete data sets.