When this option is selected, the algorithm applies user-defined calculations for certain key figures that are used to store internal machine learning regression results.

Consensus Forecast

The key figure representing the mid-term or long-term statistical forecast before any intervention by demand sensing. The algorithm makes adjustments to this forecast and optimizes it for the short-term demand. This input key figure is required to be at product-location-customer granularity level and its periodicity must be defined as Weekly.

Periodicity

Signal Horizon

The time period (past or future) from which the system considers demand signals during a demand sensing run.

Bias Horizon

Weeks in the past for which recent historical bias is calculated and used as input in the forecast bias optimization step of demand sensing. The default values mean that the forecast bias is calculated for 1, 2, 4, and 6, 8, and 9 weeks ago in a rolling manner. Specifying the bias horizon is mandatory but you can leave the last two fields empty if you wish. Make sure that you enter the values in ascending order.

Snapshot Key Figure

If your forecast model is newly created, the selection of the snapshot key figure depends on the types of the snapshot key figures added to the planning area, and on the business meaning assigned to the lag-based key figures. The following scenarios may occur:

If your forecast model was created in a release of SAP Integrated Business Planning older than 1802 and it contains a snapshot of the Change History type, the snapshot key figure is automatically selected for the model and named after the consensus demand key figure with the _REV suffix added to it.

For more information, see Key Figure Snapshots.

Maximum Forecast Increase and Maximum Forecast Increase (%)

The absolute value and percentage by which the sensed demand can be more than the consensus forecast. Both are maximum values and the higher one of the two is considered as the actual threshold. For example, if the consensus demand is 100, the maximum forecast increase is set to 50, and the maximum forecast increase percentage is set to 20, the sensed demand cannot be more than 150, which represents an increase of 50. This value is used as the maximum increase, while the specified percentage is not considered.

Maximum Forecast Decrease and Maximum Forecast Decrease (%)

The absolute value and percentage by which the sensed demand can be less than the consensus forecast. Both are maximum values and the higher one of the two is considered as the actual threshold. For example, if the consensus demand is 100, the maximum forecast decrease is set to 30, and the maximum forecast decrease percentage is set to 10, the sensed demand cannot be less than 70, which represents an decrease of 30. This value is used as the maximum decrease, while the specified percentage is not considered.

Ordered Quantity

The key figure representing the confirmed and unconfirmed quantities in sales orders and stock transport orders that are due to be delivered in the future. This non-editable key figure is always identical with the main input for forecasting steps that you choose in the Overall Parameters section.

Quantity Ratio Calculation Horizon

The length of rolling averages (in weeks) that the algorithm takes throughout the historical horizon while analyzing the ratio of total actual ordered quantities from customers to quantity of orders open at different points in the historical horizon. This information is used by the machine learning algorithm to learn from the impact of open order trends while sensing short-term future demand.

Delivered Quantity

The key figure representing the historical quantities delivered to customers.

Daily Average Calculation Horizon

The number of weeks in the past for which the requested and delivered quantities are both averaged for each day in the week. For example, a Monday average is calculated for the last 4 weeks, then a Tuesday average is calculated for the same period, and so on. The resulting daily shipment profiles are used as input for disaggregating the weekly forecasts during a demand sensing run.

Downstream Signals

Optional inputs that can help in calculating a more accurate value for the sensed demand. They provide information from sources that are between the manufacturer and the customer in the supply chain. For example, retailer point-of-sale (POS) data, warehouse inventory, and warehouse withdrawals are typical downstream signals.

It is recommended to use downstream demand signal data only when available in good quality.

You can choose up to 8 key figures for demand signals in your forecast model. The adjustments happen in the order in which these signals are added to the forecast model until the desired WMAPE threshold is reached. For more information, see Baseline WMAPE Threshold in this table.

You can only choose from those key figures that meet the following criteria:

• They have daily or weekly periodicity.

  Note Regardless of the time profile level of the downstream signal, the data is always processed on a weekly level.

• Their base planning level is Product-Location-Customer.

Minimum Data Points

The minimum number of historical data that are required for demand sensing runs.

Baseline WMAPE Threshold

The baseline threshold used by the system to check the accuracy of the consensus forecast before demand sensing is run. If the calculated weighted mean average percentage error (WMAPE) is smaller than the baseline threshold, the system does not perform the optimization steps that are parts of the demand sensing process otherwise.

This threshold is also applied when the algorithm analyzes each demand signal (forecast bias, open orders, downstream signals such as point of sales data, and so on) to optimize the sensed demand. Once the predicted forecast error is below the threshold, the algorithm stops analyzing additional signals.

Select Workdays

The days of the calendar week that the system should consider as workdays. When daily periodicity is chosen in the IBP Excel add-in, the system disaggregates the weekly consensus forecast into daily forecasts equally based on the days you select here.

Note that you must select at least one workday for this setting.

Disable Balancing and Open Order Matching

If this checkbox is selected, demand sensing will not run the balancing and open order matching steps that are a part of its logic. These steps follow a backward-forward forecast consumption logic to balance short-term demand expectations around the latest open orders and forecasts.

Default Uplift Balancing Periods

The number of weeks before and after a week with planned promotions in which high sales signals should be associated with planned promotions. This setting allows the algorithm to balance the impact of promotions across the periods where they may impact sales.

The setting is only displayed here but cannot be edited. You can edit it on the Preprocessing Steps tab after adding the promotion sales lift elimination algorithm to your forecast model.

Default Baseline Demand Balancing Periods

The number of weeks that the algorithm should consider before and after every week in the planning horizon while balancing the baseline demand by analyzing oversell-undersell patterns. Balancing can prevent double-counting of the baseline demand when orders are requested in the weeks before or after the planned consensus demand.

Default Maximum Baseline Demand Balancing %

The percentage of the consensus forecast that baseline demand balancing is allowed to consume during the “edge” weeks (that is, the weeks on the outer ends of the baseline demand balancing periods).