In order to improve performance, it is essential to measure it and make comparisons. At SKU Science, we have identified bias and absolute error as the most effective performance indicators for quickly assessing the quality of your forecasts.
When it comes to analyzing aggregated Key Performance Indicators (KPIs), there are two calculation methods available to compare forecasted values with actual demand values.
To select the preferred option for your aggregated KPIs, simply click on the gear icon located in the top right corner of the screen.
Select Rules, then KPI Computation to choose the option that interests you.
Note that the bias will remain unchanged with both methods when your data is aggregated. However, the error and thus the accuracy will provide you with different results, depending on the selected method.
Calculating KPIs and then aggregation
With the first option, the error for each aggregation is equal to the sum of the errors of the lower-level forecasts.
This is the most conservative method and is generally used by professionals for forecast analysis.
Calculating aggregation and then KPIs
The second option first sums up the forecasts for each aggregation and then compares the results to the demand for each aggregation to determine the error (and bias).
This method can be interesting for several reasons.
The first reason is that your lower-level forecasts are defined at a finer granularity level than the item level.
For example, if you have defined your forecasts at the level of item x customer, or item x distribution channel, or item x warehouse, you may be interested in visualizing the accuracy at the item level, as ultimately that's what you will produce or sell. This calculation method allows you to visualize this relevant data for your business.
Similarly, if you want to study the accuracy of your forecasts for a production machine to optimize your planning and improve your production capacity, you can obtain this information directly. Using this method, you simply need to aggregate your data by production machine, and you will directly obtain the KPIs that interest you.
Calculating forecasts using an aggregated level constitutes a second reason to use this method. Your lower-level forecasts are defined at the item level or an even finer level, but the platform calculates your forecasts at an aggregated level to achieve better accuracy, such as at the product family level.
This method allows you to directly visualize the accuracy at the chosen level for your calculation.
Let's take the last example to compare the results of the two methods.
In our case, the forecasts are calculated at the family level, and then automatically allocated to each item.
Using the first calculation method
With method 1, the sum of errors for each item results in a total average error of 17% or 607 units.
Using the second calculation method
With method 2, the error is minimized, resulting in a total average error of 14% or 477 units.
For each family, the platform displays the error obtained during the calculation.
It is interesting to note, for example, that for family 2, the error rate is the lowest at 11%. This is the family where the platform achieves the best results before the automatic allocation.
By analyzing the error for each family, you can get an idea of the accuracy of the model chosen by the platform for its calculations before distributing the forecasts to the lower levels.
Conclusion
To quickly access either of the methods, simply go to Rules and switch to the option that interests you.
By returning to the KPI analysis, you will directly obtain the data for your forecasts.