Improving the consumer demand forecast to generate more accurate suggested orders at the store-item level

Abstract

One of the biggest opportunities for this consumer goods company today is reducing retail stockouts at its Direct Store Delivery (DSD) customers via pre-selling, which represents approximately 70% of the company's total sales volume. But reducing retail stock-outs is becoming constantly more challenging with an ever-burgeoning number of SKUs due to new product introductions and packaging innovations. The main tool this consumer goods company uses to combat retail stock-outs is the pre-sell handheld, which the company provides to all field sales reps. The handheld runs proprietary software developed by this consumer goods company that creates suggested orders based on a number of factors including: * Baseline forecast (specific to store-item combination) * Seasonality effects (i.e., higher demand for products during particular seasons) * Promotional effects (i.e., lift created from sale prices) * Presence of in-store displays (i.e., more space for product than just shelf space) * Weekday effects (i.e., selling more on weekends when most people shop) * Holiday effects (i.e., higher demand for products at holidays) * Inventory levels on the shelves and in the back room * In-transit orders (i.e., orders that may already be on their way to the customer) The more accurate that the suggested orders are, the fewer retail stock-outs will occur. This project seeks to increase the accuracy of the consumer demand forecast, and ultimately the suggested orders, by improving the baseline forecast and accounting for the effect of cannibalization on demand.