In this paper we develop a forecasting algorithm for recurrent patterns in consumer demand. We study this problem in two different settings: pull and push models. We discuss several features of the algorithm concerning sampling, periodic approximation, denoising and forecasting.