Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a mathematical model to predict whether a future software update will be applied to one of the one or more computing devices using information collected from the devices, the collected information including the following for each of a plurality of software updates made available during the one or more time periods: an identification of software to which the update applies, whether the software update was installed on any of the computing devices, and, if so, whether the software installation was successful.