Pega AI is primarily based on self-learning models Reference:

Pega AI is primarily based on  self-learning models .

According to the Pega Academy 1 , a strategy is a unit of reasoning that defines how to select an action for a customer. A strategy can be defined in different issues and groups, which are categories that help organize actions. An issue represents a business goal (such as retention or sales), and a group represents a subcategory of an issue (such as cross-sell or up-sell).

In a strategy defined in the “Retention” issue and the “X-Sell” group, you can import  actions from all groups under the “Retention” issue 1 . This allows you to use actions that are relevant to your business goal and compare them with other actions in the same issue.

The performance of an adaptive model that has not collected any evidence yet is 50. This means that the model is not confident about its predictions and assigns equal probability to all actions. References: https://community.pega.com/sites/default/files/help_v82/procomhelpmain.htm#rule-/rule-decision-/rule-decision-adaptivemodel/main.htm

The pyPropensity property stores the likelihood that an action will be accepted by the customer. It is calculated by a predictive model or an adaptive model and used in decision strategies to prioritize actions. References: https://academy.pega.com/module/creating-and-understanding-decision-strategies-archived/topic/using-predictions-decision

To predict if a customer is likely to churn, you use a model of type decision tree. A decision tree is a type of predictive model that uses a set of rules to classify customers into different categories based on their attributes and behavior. A decision tree can predict a binary outcome (such as churn or not churn) or a multi-class outcome (such as low risk, medium risk, or high risk). References: https://academy.pega.com/module/predictive-analytics/topic/using-decision-tree-models

To offer a 10% discount for customers whose CLV value is higher than 400, you should use the  Filter  strategy component.

A solid arrow from a “Set Property” component to a “Filter” component means that a property from the “Set Property” component is referenced by the “Filter” component. For example, you can use a “Set Property” component to calculate a customer’s age and then use a “Filter” component to remove actions that are not suitable for that age group. References: https://academy.pega.com/module/creating-and-understanding-decision-strategies-archived/topic/setting-properties

Adaptive model predictors are selected from the customer profile, which contains information about the customer’s attributes and behavior. Predictors can be either scalar or aggregate properties that capture customer context, such as channel, location, time, etc. References: https://academy.pega.com/module/predicting-customer-behavior-using-real-time-data-archived/topic/configuring-adaptive-models

 When you place the new model in shadow mode in the production environment, the current model still drives the prediction, but the new model runs in parallel and collects performance data for comparison. References: https://academy.pega.com/module/predictive-analytics/topic/mlops

To comply with the new business rule of showing a proactive retention offer to high churn risk customers, you should modify  the Retention issue .