The Prediction Performance found in the Model Summary tab, is a measurement of how well the model is able to predict your historical performance. If your model isn’t capable of correctly predicting what happened in the past, it shouldn’t be trusted to predict future performance.
Using your historical spend, we ask the model to predict your target variable (something like revenue) for every day in the past. Because we are using historical data, we know how much revenue you actually generated each day as well. The Prediction Performance metric expresses how close the predicted revenue is to your historical revenue - 100% means a perfect match.
You may see a warning that says “Your Model Is Performing Below Benchmark” when viewing your predictions. This warning occurs when the Prediction Performance for your model is lower than 65%. In this case, we don’t feel confident in the recommendations being made by the model.
This can occur for a variety of reasons: non-marketing forces driving your target variable, outliers/gaps in your data that require remediation, etc. If you see this warning, you should reach out to your CSM to discuss improving your model performance.