Demand Forecasting Across Industries

"Demand forecasting" is not a monolith. The algorithmic choices, data dependencies, and mathematical loss functions fundamentally change depending on the physical nature of the product and its economic lifecycle.

This article contrasts four highly distinct domains: Fresh Food, Fashion, Consumer Electronics, and Heavy Machinery (Spare Parts).


1. Fresh Food: The Battle Against Physical Decay


2. Fashion & Apparel: Trend Volatility and Markdown Optimization


3. Consumer Electronics: Cannibalization and Hierarchies


4. Heavy Machinery & Spare Parts: Intermittent Demand


Summary Matrix

IndustryPrimary ConstraintKey Algorithm / ApproachForecasting Horizon
Fresh FoodPhysical SpoilageAsymmetric Loss (Quantile Regression)Intra-day / Daily
FashionTrend ObsolescenceAttribute Clustering & Markdown ElasticitySeasonal
ElectronicsComponent ComplexityHierarchical Reconciliation & CannibalizationMid-Term (Months)
Spare PartsIntermittent DemandCroston's Method / Installed BaseLong-Term (Years)