Reduce Errors by
Automating Your
Forecast Model
Selection

  • by Dmitriy Mindich, Senior Associate, PricewaterhouseCoopers, LLC
  • December 15, 2008
Learn how to use the Auto 2 algorithm in SAP Advanced Planning and Optimization Demand Planning to automate your statistical forecast generation process.
Key Concept

A statistical forecast is the backbone of the consensus demand plan process. Even though most experts agree that statistical forecasts always contain forecast errors, the forecasts provide an invaluable baseline for comparison against all other forward-looking forecasts (marketing, sales, brand, or customer). The Auto 2 algorithm of SAP Advanced Planning and Optimization Demand Planning can be used to automatically generate a forecast for a large number of stock keeping units (SKUs).

When using the SAP Advanced Planning and Optimization (APO) Demand Planning (DP) module, companies are faced with a daunting task of selecting the right forecasting models. While each company's business is unique, most have one thing in common: heterogeneity of product offerings. From the perspective of Demand Planning, this “commonality” entails seasonality, trend, or intermittent patterns in history that should be identified for every product, product group and exploited for better forecasting results.

To begin forecasting, a demand planner typically goes through the process of choosing a unique forecasting model for a designated selection of products. The selection of products and the use of a unique forecasting model assumes homogeneity in the demand history across all products in the specified selection. Unfortunately, this assumption can result in large accuracy gaps across all forecasts generated. Essentially, the same forecast model is force-fitted across multiple products.

SAP APO Demand Planning offers a variety of forecasting algorithms used to generate a statistical forecast. The Automatic Model Selection 2 (Auto 2) algorithm, SAP's version of the pick-best approach, provides for a way to automatically select a forecasting model for each subset of products in the selection. Furthermore, while the system minimizes various forecast fit errors, such as the Mean Absolute Percentage Error (MAPE) and the Mean Absolute Deviation (MAD), it will also identify the necessary parameters of the forecasting algorithm with minimal user interference.

Dmitriy Mindich

Dmitriy Mindich, APICS CSCP, is a senior associate at PricewaterhouseCoopers LLP. He has extensive SAP experience and specializes in SAP SCM areas of SAP APO Demand Planning and SAP APO Supply Network Planning, as well as integration with external systems using the SAP NetWeaver Process Integration platform.

See more by this author


Comments

No comments have been submitted on this article. 


Please log in to post a comment.

To learn more about subscription access to premium content, click here.