Demand Forecasting

What it covers

The order follows a real implementation: first what you forecast, then how you measure it, and only at the end what you do with the number.

1. FundamentalsWhat you forecast and what you do not. Sales versus demand. Granularity, horizon and time aggregation. How forecasting differs from planning.
2. Forecast accuracyMAE, RMSE, MAPE, WAPE and bias. What each metric does when you change the aggregation level, and why one number on its own says nothing.
3. BaselinesNaive, seasonal naive, moving averages. What a baseline is for, and how to tell whether the complex model is adding anything.
4. Statistical forecastingTrend and seasonality. ETS, and the ideas behind ARIMA. When a statistical model is more than enough.
5. Machine learning forecastingThe problem framed as supervised learning: lags, rolling statistics, calendar features. Global models, tree-based models, backtesting and selection.
6. Foundation modelsModels pretrained on millions of series that forecast one they have never seen. What they actually solve, how to evaluate them without swallowing benchmark contamination, and what they cost against one trained in house.
7. Promotions and external signalsPromotions, price, holidays and events. Known-future versus unknown-future variables, and the most common ways of leaking the future without noticing.
8. Stockouts and censored demandSales that are not demand. Detecting constrained periods, estimating what was lost, and what changes in the model.
9. New and intermittent productsStarting with no history. Analog products, sparse series, intermittent demand, and the approaches that combine classification with regression.
10. Probabilistic forecastingFrom the point estimate to the distribution. Quantiles, intervals, and why every inventory decision needs the uncertainty and not the average.
11. Hierarchical forecastingSKU, location, category. Bottom-up, top-down and reconciliation, and why the category number almost never equals the sum of its parts.
12. Human overridesWhy planners override, when the override adds value, and how to measure it. Traceability, and the loop between the person and the model.
13. From forecast to decisionLead-time demand, safety stock, service level and replenishment. What overstock costs against what a stockout costs.