AbstractIn a service environment, a stockist usually has many slow moving items whose infrequency of demand can give rise to forecasting problems. Moreover, when a demand occurs, the request is sometimes for more than a single unit, which results in so-called lumpy demand. In this paper, the standard method for dealing with such intermittent demand is reassessed. Some general results are presented that enable variance estimates to be made, and these are particularly straightforward when the demand occasions can be represented as a Poisson process. Some experimental evidence is advanced to support this model in the specific situation under study. Since EWMA forecasts are central to many commercial systems, a simulation analysis was conducted to determine under what conditions intermittent demand requires its own model, rather than an unadjusted EWMA. Superior performance is demonstrated for items that have an average inter-order interval greater than 1.25 forecast review periods, and the magnitude of the improvement increases as the average interval lengthens.