Paper 1180-2021
Authors
Ajay Kumar Mishra, Ph.D.; Mohak Narkhede; Chandu Saladi; Core Compete
Abstract
A leading e-commerce retailer with multiple websites desired to create a systematic, scalable and science-driven product forecasting system. SAS Forecast Studio was configured on a cloud-based server for automated weekly forecasting of tens of thousands of products for the next six months. SAS was used for data insights, such as understanding distribution of product life and rate of sale, variability in demand and prices, as well as seasonality among the products. Pre-processing included creating demand history by adjusting order history for lost sales using inventory data. Price is an important predictor of demand for this retailer. We modeled holiday events using HPFEVENTS in SAS Forecast Server procedures. With a deep understanding of the patterns in the data, SAS Forecast Studio was configured for test runs to select independent variable form, fit criteria, reconciliation and hierarchy choices. The heterogeneity of the patterns ensured that ARIMA and UCM with predictor variables, as well as ESM, intermittent demand models and combined models emerged as winning models for a significant portion of products. In addition to several standard reports, elaborate accuracy reports were created, including forecast value added measurements. Steps were also added to identify exception forecasts and replace or modify them. The SAS Forecast Studio run was automated using an FSCREATE macro.
See attached for all the details.
Watch the presentation
Watch Large-Scale Demand Forecasting for E-Commerce Retailer Using SAS® Forecast Studio: A Case Study as presented by the authors on the SAS Users YouTube channel.
