A method for determining machine usage severity including collecting repair data for multiple machines over a repair time period and telematics data from sensors on the machines over an activity time period. The method can include calculating predictive features from the telematics data for each of the machines and creating a severity model based on the predictive features. The severity model can be validated with the predictive features and corresponding repair data for each of the machines. The method can include receiving telematics data from sensors on a deployed machine for a deployed period of time, calculating a plurality of machine predictive features from the telematics data, and feeding the machine predictive features into the severity model to calculate a severity score for the deployed machine. The method can include displaying a recommendation to perform maintenance on the deployed machine when the machine usage severity score exceeds a selected threshold.