This article describes a operational research problem. A company that specializes in temperature controlled transportation wants optimize the planning of its human and material resources through short term activity forecasting. The challenge is to find a unique forecasting model adapted, without human intervention, to the specific needs of 57 company's officies. To do it, the company collected data since five years. To analyse them, mathematical algorithms for forecating time series are used. The work is to combine these tools to extract the maximum of determinist information that should be anticipated. The introduction presents the problem and its economic context. This is followed by a description of the process used and arguments to defend choices done. The adopted solutions are inventoried. Finally, the conclusion refers to courses of study.