SAP IBP's demand module incorporates advanced AI and classical machine learning features to enhance forecasting accuracy. Machine learning applications are used in pre-processing steps and within the statistical forecasting algorithm library.
Your product's sales behavior isn't static. How do you ensure you're using the right forecasting model when underlying trends or demand patterns shift unexpectedly?
Segmentation and Time Series Analysis:
Processes such as ABC/XYZ segmentation group products by their properties, allowing for the selection of the most suitable forecasting method.
K-means clustering algorithms define segments automatically after the user selects the segmentation measure and desired number of segments. Segment thresholds are dynamically adjusted as data changes.
Change Point Detection:
Time series analysis detects whether data is continuous, intermittent, or has special trends/seasonality. Classical statistical methods are applied, with machine learning-based change point detection identifying significant shifts or structural breaks, such as level or trend changes from new sales or market shifts.
Detected change points can inform forecasting models, including multiple linear regression, ARIMAX/SARIMAX, and gradient boosting decision trees. This results in more stable forecasts and reduces manual intervention.
A single data anomaly, like a one-time bulk order, can throw off your entire forecast. How can you confidently and automatically clean your historical data without tedious manual intervention?
Automatic Outlier Correction with Voting Mechanism:
Outlier detection corrects anomalies in transactional data, combining multiple algorithms (variance, interquartile range, machine learning methods like MED, DBSCAN, and isolation forest). A data point is marked as an outlier only if the majority of methods agree, increasing robustness and automation.