MA.5:4.2 - Construct compatible demand inputs

Fix the output being forecast, the population, period and information cutoff. A forecast of orders accepted next month differs from one of units delivered next month: carry the order backlog and promised delivery dates into the latter. Distinguish units, customers and transactions before combining them. MA.8 supplies the time and population account when several customer cohorts or product phases contribute.

Separate what is already included from what remains to be estimated. For a delivery forecast, begin with due work expected to survive cancellation or rescheduling, then estimate further work that can enter and be delivered in that period. Keep those components disjoint. If a supplied forecast already covers all deliveries, adding current orders to it would count part of the population twice.

Construct the missing component from evidence suited to it. Comparable history can supply a simple benchmark, such as the average further orders received after the same cutoff when demand conditions are stable. Compare like populations and horizons; a seasonal series needs a relevant seasonal comparison, and a changed offer or sales channel can defeat an old average. Judgment can incorporate information absent from that history, but name the changed condition and how it affects the quantity. A specialist forecast can be used directly when its population, cutoff, horizon and assumptions fit.

Begin with the simplest supported forecast that can change the decision usefully. Compare added complexity against that benchmark on observations not used to fit or choose the model when predictive accuracy is the claimed gain. Obtain a forecasting method or specialist contribution where estimation requires it. A few teaching observations, an attractive fit or three labeled scenarios do not establish predictive calibration.

Keep observed facts, commitments, estimates and proposed actions distinguishable. State whether a precise statistical claim reports a mean, quantile or another defined quantity. A central planning scenario is not automatically an expected value. Where the evidence supports only a conditional scenario, preserve that qualification instead of presenting the scenario as an evidenced expectation.