The forecast estimates the potential demand for the future. Demand Forecasting uses several calculation methods.
The app detects the best method for your data, but you can also select from the following forecast methods:
- Naive forecast
- Moving average
- Exponential smoothing
Forecast graph and table¶
The forecast is calculated individually for each record, and for each level.
The graph and table represents the following results of the calculation:
- Historical data (Cleaned or raw, depending on your choice)
- Proposed forecast
- Base forecast
- Top down/Bottom up forecast
The graph and table show data for the record selected in View side panel tab. The selected record is named in the header of the graph and table.
You can edit the Proposed forecast by double-clicking a value in the table. Other data rows cannot be edited here.
Changes in a record’s data for a period are highlighted, and proposed forecast is recalculated on other levels by aggregation/disaggregation.
Action on the table
To compare the results of the forecast, the forecast for each level is calculated top down and/or bottom up. You have these options in the action menu in the top right corner of the table:
- Mark as reviewed: the selected records is marked as reviewed
- Set to base: the proposed forecast for the selected record is set to the base forecast
- Set to top down: the proposed forecast for the selected record is set to the top down forecast, available on Low and Medium level
- Set to bottom up: the proposed forecast for the selected record is set to the bottom up forecast, available on Medium and High level
- Force down: the proposed forecast for the highest level is set to force down forecast
Arriving to the Forecast Review page¶
You are automatically directed to this page if:
- you select Approve all & calculate forecast on the Data cleaning page, and select to view results.
- you change the forecast settings on Forecast methods page and recalculate.
You can also navigate to this page from 2nd step of Workflow or from the top menu Forecast > Forecast Results.
If there is historical data not yet approved, a dialog asks if you want to view the previous results, or calculate the forecast based on raw data.
The Forecast settings widget shows the forecast method and values for each selected record. You can change the selected forecast method, or the forecasts’ parameter values.
To change multiple forecast methods for different levels, go to the Forecast methods page.
You can see the current state data in the following places:
- Forecast Graph and Table title
- Status bar: Here you can see aggregation level and current record’s number in the record list.
- View side panel: Here you can see the current aggregation level and record
You can filter the record list via Filters side panel. When the selected filter is turned on, the Record list is filtered automatically and Previous!Next button got the ‘filtered’ word.
You can filter the forecast data based on the historical period.
- Filter records where historical period numbers are less than a defined one.
- Filter records where periods with zero demand are more than a defined number.
Adjust tracking signal on the Period Length side panel. There are fields for the total historical period number and the set forecast history length in periods. Below is a table to set the period numbers.
Changed Forecast Method
You can filter records based on the forecast method has changed by the learning algorithm.
Only Reviewed records
You can filter those records which are already reviewed.
Recalculate selected record¶
This button becomes active when you change the Forecast settings. The proposed forecast will be updated for all levels by aggregation/disaggregation.
Approve forecast changes¶
Click this button to confirm the proposed forecast. A dialog asks if you to stay on this page, where you can review the changes, or export the results, which opens the Export results page.
From here, you can go to Export results through the Workflow to download the forecast data, or go to the Forecast methods page to change multiple forecast methods for different levels and recalculate.