Forecasting of customer demand for seasonal goods using retail curve vector

Abstract

A method for forecasting the demand for seasonal goods using the vector of distribution of sales during the year, called retail curve vector is proposed. Components of retail curve vector are the weekly sales volumes of the considered or similar product, obtained on the basis of sales statistics for the previous calendar year. The condition of applicability of the proposed method is the fulfillment of the hypothesis about the convergence of the corresponding weekly sales volumes of two consecutive years and the hypothesis about the existence of goods groups with similar sales dynamics. The application of the method allows to build a demand forecast for the goods within the next week based on sales volumes data for the previous few weeks of the current and previous years, as well as sales volumes data for the same week of the previous year. Software architecture to implement proposed method for forecasting demand using a micro-service architecture based on the Google Cloud Platform is presented. Such components as Google Kubernetes Engine, Google BigQuery, Redis as used. To reduce the computational load on the main system, necessary data is copied to the OLAP system and required forecast is build without usage of the OLTP system. The results of numerical experiment on forecasting the demand for goods, obtained on the basis of real data, are presented. Comparison of the results of demand forecasting using the retail curve vector and the moving average method is performed. The possibility of using proposed method of demand forecasting as a component of an automated inventory control system in supply networks is shown.