Forecast modeling

Demand Planner reviews a model assignment for each planning entity before using its output. The model record states the method and its parameters, whether it is active, when it was assigned and who assigned it. A run then records the date, run type, history used, periods generated and error measures. Keep the model and run connected to the planning entity so a baseline quantity can be traced to its source decision.

Forecasts hold period amounts for the baseline, adjusted, consensus and final positions. Those amounts answer different questions. The baseline shows the statistical starting point. The later amounts hold commercial and reviewed changes without erasing that starting point. Actual quantity allows the next cycle to compare the published judgment with demand observed later.

• Select the method. Use the active model assignment and its recorded method for a specific planning entity.
• Read the run. Compare MAPE, WAPE and bias for a run, along with its history depth and generated horizon.
• Inspect seasonality. Review the seasonal profile’s method, strength and ordered index points before relying on a repeating pattern.

The team should agree how much history is sufficient for each item class, when a manual override is acceptable and who reviews a model reassignment. Define how the accuracy threshold is used during review. A low error measure from a short or unrepresentative history should not become a final plan without context. Check that drivers such as price, weather or marketing spend have a period, value, unit and source that match the forecast series.

A regression model can point to driver records through its driver list. This lets the planner explain what changed in a model rather than attributing every movement to seasonality. Forecast error metrics also retain error measures by cycle, planning entity and version. Compare those with run-level measures: one describes the model execution, while the other supports cycle review.

Consider a retail item with a clear holiday peak. Demand Planner compares an active seasonal model with the last run and the profile’s monthly points. The planner checks that the run used enough history and that bias is not persistently over the actual quantity. Then the planner opens the period forecast and identifies the baseline amount linked to that run. When a later commercial adjustment changes the adjusted amount, the baseline remains available for comparison.

If the forecast starts drifting, record an exception for the affected period and assign an investigator. Review the prior runs before changing the assigned model. Keep the model change dated so the next reviewer understands which version produced which baseline.

• Model run accuracy — Run method context, date and error measures by planning entity.
• Forecast worklist — Period amounts side by side, including actual quantity when available.

Use baseline preparation before commercial adjustment review.