The retail companies are constantly seeking to increase their competitiveness. Among the factors that influence this competitiveness, this thesis focuses on the cost of supplies and the level of customer service. The level of service depends on the planning of inventories, which aims to balance the supply of products to the demand. The starting point of this planning is the demand forecast. Thus, this thesis proposes a routine of monitoring forecasting errors using statistics called Tracking Signal. These statistics are used to detect when the forecasting error out shows the expected behavior, thus indicating a bias in the forecasts. For this purpose, an experiment was conducted which resulted: determination of detection limits for these statistics; identification those most effective in detecting bias in forecasts by simple exponential smoothing. Moreover, the reduction in costs of supplies and improving the level of service are also covered by this thesis proposed an inventory model. This model considers an item on a system with two echelons and periodic review, consisting of a Warehouse (W) and N retail stores. The retails revisions apply in equal and synchronized interval; it is possible the W partially deliver the orders to the stores in case of shortages. For this model, a search algorithm based on the golden section method, which minimizes the total cost in the system given the target level of service in retail was developed.