We present PerForecast, a tool that aims at making the process of capacity planning automatic, by using univariate time series datasets to make predictions. Those predictions are based on classical predictive models. They are all compared to select the best one – the one we display to the customer. The objective is to anticipate the sizing of Orange infrastructures, in order to predict the power loss of a server, and to order new machines in advance. Indeed, those procedures are long and costly, and the sooner the better for our Quality of Services (QoS).