Use sales history to measure demand for forecasting, not item requests or demand data. Adjust sales history to exclude unusual orders. Bodenstab recommends an automatic filter on sales history of four times the mean average deviation between the forecast and actual usage (explained below).
Graham offered two forecast methods, one for non-seasonal items and one for season items. For non-seasonal items, use a simple average of the usage for the last six months. For seasonal items, use a simple average of the next six months from a year ago and adjust the number with an estimate about how this year will vary from last year. Bodenstab and Schreibfeder expand simple averages with weighted averages (or exponential averages) to give greater weight to more recent or to more seasonally-relevant demand. Bodenstab continually refines the forecast by capturing the forecast error each month to calculate the actual mean average deviation. The result is also used to filter sales history automatically.
How Much to Order?
Without consideration of other factors, the optimal inventory level is achieved by ordering as often and as little as possible. An order is placed every day to cover demand the day the inventory arrives. Even for non-seasonal items, however, there are other factors to consider. Demand is not constant. There are costs associated with processing orders. Vendors sell only in case or pallet quantities or have minimum order quantities for a total order. Then there are tiered price breaks or the best price or freight rate for combined orders that fill a truck or container. As a result of these and other factors, producing a good demand forecast is only the beginning of calculating an optimal order quantity.