How to measure the accuracy of your network?

The RetailSonar predictive model generates a performance prediction.

It’s important to assess the accuracy of the model for your network, as this reflects the overall deviation between predicted and actual performance for your current network. Model accuracy ofcourse says something about the way it can be applied to future network changes.

When no location characteristics are provided, you can expect an accuracy of 75–80%.
When all location characteristics are correctly configured, the accuracy generally increases to 80–85%.
However, the final accuracy also depends on factors such as brand consistency and the overall strength of the store concept.

This article explains how to use the location performance framework to measure the accuracy of your network.

Remark! If the sales data has already been uploaded to the platform by the Power User, you can skip steps 5 and 6. For more information on how to upload sales data to the platform, click here.

When you receive your training on the solution, you will also receive a template document for calculating the accuracy of the solution on your network. The article explains how you can do this yourself by making an export from the location performance framework. 

1. Open the Location Performance Framework
2. Load the default framework
3. Export the framework for further analysis in excel
   
   
    
4. Open the exported location performance framework in Excel.
   
   
    
5. Upload the actual sales data into the Real Sales column. This step is not necessary if your sales data is already uploaded in the application. For more information on how to upload sales data directly into the platform, click here.
6. Calculate the deviation between the predicted and actual sales in the Deviation column using the following formula:
   Deviation = (Benchmark - Real Sales) / Real Sales. This is already implemented in the template that RetailSonar provides.
7. Identify and evaluate the biggest outliers. Ensure that all location characteristics are filled out correctly.
   If the location characteristics are accurate, investigate whether these outliers are locations you're familiar with. For example, have there been construction projects throughout the year, or is there exceptional management that could affect performance?
   If there are exceptional factors at a location that influence the performance, exclude it from your network when analyzing accuracy. These deviations should not be included in the overall accuracy measurement, as the model cannot correct for these conditions
8. Identify the brand strength. RetailSonar sets up an application with the best knowledge out there. That means that it is possible that there is some refinement to make in your overall brand strength. It is thus possible that the overall estimation is generally too low or too high. This will ofcourse impact the accuracy. The template that is provided to you will allow you to correct this general brand strength.
9. Measure the overall accuracy of your network by evaluating the deviation of all locations in scope. Use the following formula to calculate the accuracy:
   Accuracy of the network = 1 - (SUM (ABS(Benchmark - Real Sales)) / SUM(Real Sales))

Remark! It is important to take the absolute difference between the benchmark and the real sales