Improving automotive garage operations by categorical forecasts using a large number of variables

•We develop a holistic framework for decision making in automotive garages.•We focus on the categorical prediction of repair times, rather than point forecasts.•We consider a large number of predictor variables (with condition, manufacturing, geographical, and calendar-related information) coupled with suitable variable selection techniques, namely LASSO and the elastic net.•We further link the predicted probabilities to operational decision making in garages.•We reveal that best decisions do not always correspond to point forecasts, especially under asymmetric loss functions.