Demand
Structure and methods
Demand management covers forecasting, communicating the plan, influencing demand, and allocating supply when it is short. A forecast decomposes into level, trend, seasonality, cyclical, and random components. Qualitative methods (sales force composite, Delphi, market research, panels) suit new products and long horizons; quantitative methods (moving average, exponential smoothing, decomposition, regression, simulation) suit items with history.
Metrics that appear on the exam
Forecast error is managed, not eliminated. Bias is directional and is tracked by RSFE; the tracking signal (RSFE ÷ MAD) flags when a forecast has drifted out of control. MAD is the average absolute error and drives safety stock sizing; MAPE expresses error as a percentage but breaks down near zero demand. Forecast value added compares a forecasting step against a naive benchmark - if it cannot beat last period's actual, drop the step.
Traps
- Confusing dependent demand (components, calculated from the MPS) with independent demand (end items, forecast).
- Believing a high smoothing constant is more stable - high alpha tracks recent change; low alpha is stable.
- Applying stale seasonal indices to a product whose trend has changed.
- Chasing accuracy at the wrong aggregation: accuracy improves with aggregation and horizon, but the schedule needs detail. Pyramid forecasting reconciles top-down and bottom-up.
- Treating a good forecast as good plan attainment.
Worked example
Exponential smoothing: F(t+1) = alpha × A(t) + (1 − alpha) × F(t). With alpha = 0.3, actual 120, prior forecast 100: 0.3(120) + 0.7(100) = 36 + 70 = 106. Rule of thumb: select alpha by minimizing MAD over a representative history, never by intuition.