Hierarchy and DFUs

A DFU (Demand Forecast Unit) is the smallest combination of dimensions a forecast is produced for. After upload, the Hierarchy and DFU step lets you choose those dimensions and their order.

Dimension order

The list order is the hierarchy order. Drag or move columns to change it.

• The combination of selected values identifies one DFU. With family, product, and location, each DFU looks like Beverages/Coffee/North.
• The order also sets aggregation levels used later for reconciliation: lower levels should add up to the levels above them.

Select at least one dimension. date and demand are fixed and cannot join the hierarchy. The same dimension cannot be selected twice.

What is not in the hierarchy

Every remaining column is evaluated as a regressor — an external signal such as price or promotion. You do not pick regressors one by one in this step; leaving a column out of the hierarchy is enough.

If a column is an identifier you need in the forecast grain, put it in the hierarchy. If it is an explanatory signal, leave it out.

Preview before you continue

The page shows a sample of the uploaded columns. Check that:

• hierarchy values are the business keys you plan with;
• numeric measure-like columns are not used as dimensions unless that is intentional;
• leftover columns are signals you actually want the models to see.

Validate and analyze submits the hierarchy, starts validation, and opens the Analysis step. If the uploaded file cannot be inspected, restart the upload from Data rather than editing the hierarchy.