Demand Planning¶

Demand Planning combines statistical evidence and business knowledge into one traceable demand signal. The goal is not only to generate a forecast, but to understand which assumptions changed it, who contributed, and whether each stage improved the result.

What the process produces¶

The journey separates objects with different purposes:

• Simulation tests a model and its parameters without creating the official plan.
• AutoFit compares candidate configurations through historical backtests and recommends the best fit for the selected objective.
• Demand Plan is the persisted, versioned plan.
• Planning Book is the workspace where teams analyze and collaborate on that plan.
• Released Demand records which Demand Plan was official for each sales period, preserving a fair basis for retrospective accuracy.

This separation lets a team experiment freely while keeping the official baseline, human changes, and released plan auditable.

Forecast level and demand behavior¶

The forecast does not need to run at the lowest product-customer level. A team can model demand at a more stable cluster and then distribute the result to the operating detail. Clusters may represent product families, channels, customers, regions, lifecycle stages, or combinations of planning characteristics.

This matters because different series behave differently. A stable family should not be forced through the same treatment as a sparse new product, a seasonal channel, or an intermittent customer. The correct level balances statistical signal, explainability, and the detail needed by downstream planning.

See forecast levels and clusters and the forecast cluster contract.

Statistical baseline, external series, and AI¶

The Community edition provides the transparent statistical baseline and the basic workflow needed to create a Demand Plan. Enterprise extends this foundation with advanced statistical models, AI models, external support series, richer history treatment, and advanced disaggregation.

An external series can represent a causal signal such as price, an index, weather, traffic, or another business driver. Enterprise models may use that signal to improve the aggregate projection while preserving the individual demand profile of each product, customer, or material-location during disaggregation. The result is not a flat proportional split: the operating detail can retain its own cadence, seasonality, and mix.

Use cluster-level AutoFit to understand the initial supported configuration.

Configurable collaboration by layers¶

A plan can move through any sequence of configured stages, for example:

1. statistical baseline;
2. Marketing collaboration for campaigns and events;
3. Engineering collaboration for introductions and transitions;
4. Sales collaboration by supervisors;
5. Sales collaboration by managers;
6. consensus approval and release.

Each stage can have its own scope, responsibility, editable key figures, and approval rule. Enterprise can preserve the value contributed by each stage and compare its accuracy with the statistical baseline and later stages. This makes it possible to identify where collaboration improves the plan and where it systematically introduces bias or destroys value.

See configure the collaboration workflow, collaborate, and measure forecast accuracy.

Recommended journey¶

1. Prepare master and sales data.
2. Define forecast levels and clusters.
3. Configure cluster-level AutoFit.
4. Run Demand Planning.
5. Analyze the Demand Plan.
6. Configure the collaboration workflow.
7. Collaborate on the plan.
8. Measure accuracy against the plan that was official for each period.

Data map¶

The forecast lives at the intersection of material, location, and time. See the complete Demand Planning data model.

• Materials define what is planned.
• Locations define where demand is observed or served.
• Units of measure make quantities comparable.
• Sell-out represents sales observed at a location when identifying the final destination is not relevant or possible.
• Sell-in represents sales where both the internal origin and the customer or destination are known.

Both sources may coexist. In collaborative planning with a distributor, the supplier records sell-in to the distributor; the distributor may share sell-out and inventory; and the projected sell-out, combined with projected inventory, supports the future sell-in needed for replenishment.

Readiness and control¶

A process is ready when the team can explain the historical source and last closed date, forecast level, disaggregation rule, planning unit, horizon, cluster ownership, collaboration stages, downstream consumption, and the released plan used for accuracy. Begin with a reconciled scope and expand only after the team can explain how a series entered the plan.

Edition comparison¶

The edition boundary follows executable capabilities, not labels shown in a screen. See the Community overview for the versioned public baseline.