Enhancing Net Sales Forecasting by Applying a Clustering-Based Hierarchical Method

Accurate net sales forecasting is crucial for a company’s planning. At different hierarchical levels, cluster analysis is useful for solving the trade-off between bottom-up and top-down forecasting approaches. We develop a hierarchical method to enhance accuracy at different hierarchical levels. We create clusters based on statistical time series features rather than natural attributes such as geography, customer type, or product categories. Emphasizing a staged research process with “build” and “evaluate” activities, we choose a science and technology company as our case. Our approach provides business professionals with an understanding of the applied algorithm and the potential for expert recommendations. Product launches, withdrawals, high fluctuation, sales with similar seasonality, upward/downward trend sales, etc., can be automatically clustered. At the case company, we improved the forecasting performance significantly compared to the linear regression, seasonal naïve, SARIMA and Holt-Winters methods currently applied without clustering.