• A lot of offers are being rejected which indicates a problem with the underlying models
• Response timeout is enabled, but explicit responses are not coming in within the timeout period, which leads to implicit negative responses.

• A lot of offers are being rejected, which has decreased the success rate for the test group, which also indicates a problem with the underlying models.
• The control group is not representative enough of the actual population of the subjects or customers.

• The outcome-based model has not received enough responses with positive outcomes.
• The model does not have the right set of predictors.
• The actions available to the model are too restrictive.

In a model update scenario, in which a shadow model runs alongside an active model, the performance chart displays the performance curves for each model in the active-shadow pair. The comparison can help you determine which model is more effective and decide whether you want to promote the shadow model to the active model position. The analysis is available for both outcome-based and supporting models.

The following figure shows the performance chart for the active model, Predict Churn DRF. This model is paired with a shadow model, Churn GBM. The chart shows two distinct lines that illustrate the performance of each model over time. Hover over the lines with your mouse pointer to see the exact scores for each model at different times. In this example, the two lines converge around January 27, suggesting that the models performed equally well. The detailed information that is displayed in the tooltip shows that the active model has a slightly higher score.

• Limited actions or offers are introduced that are not receiving any responses.
• The response strategy is configured incorrectly and the Decision Architect needs to review it.

• The predictor configuration is wrong.
• The model is not learning from responses.
• The actions available to the model are too restrictive..
• The model was recently deployed and has not received responses yet. The propensity is smoothed by around 0.5, by adding more noise.
• Positive responses are declined in the course of time which could be because of:
  • Limited actions or offers are introduced.
  • The predictor configuration is wrong.

• The mapping of predictors is incorrect.
• The Pega Platform property mapped to the predictor is not getting the value from an underlying source.

• The symbolic predictor is not sourced properly from the corresponding data source.
• The configuration of predictors in the strategy is incorrect. When parameterized predictors are used, ensure that the model is receiving the inputs properly with the Supply data via model component.
• There is a change in inflow of symbols for the predictor from the data source.

• The predictor may be sourced incorrectly from the underlying Pega Platform data model.
• The mapping between the Pega Platform data model and the predictive or adaptive model may be incorrect.

• The actual predictor values have changed.
• Symbols appearing or disappearing over the selected time frame can indicate the addition or deletion of symbol levels that could affect model performance.
• The class property is not getting the correct values from the prediction strategy or model.

• There has been data drift for the models.
• The model has received too many missing values, which has resulted in percentile drift.

• There has been data drift for the models.
• The model has received too many missing values, which has resulted in percentile drift.

• Predictor mapping inside the model or strategy is incorrect.
• The predictor value is not available from the corresponding data sources or not mapped correctly to the Pega Platform property.

• Predictor mapping inside the model or strategy is incorrect.
• The predictor value is not available from the corresponding data sources or not mapped correctly to the Pega Platform property.