Automated time series demand forecast for luxury fashion online retail company

Abstract

Demand forecasting for a retail company in luxury fashion is a challenging process due to the highly complex and demanding customer profile. As the company keep growing, more and more partners are demanding the expected volume of orders for better operational capacity planning and to justify the return of their investment. This project aims to create an automatic and scalable forecasting process to ensure customer experience and partnership profitability. By studying decomposition time series forecasting taking in consideration the customer behavior, a machine learning process can be applied for parameters tuning depending on customer clusters based on geolocation and marketing events. The proposed process has shown forecast accuracy number up to 90% for non-sale season and 84% for sale season periods, reducing the forecasting time in 88% versus the previous forecast process and increasing the partner coverage from 20% to 100%. Acknowledging that this forecast process is a continuous learning process, the foundation of a robust supply chain planning was created building trust in the organization and adding value to the partners.