Skip to main content

         This documentation site is for previous versions. Visit our new documentation site for current releases.      

Creating predictions for customer engagement

Updated on September 15, 2022

Optimize customer engagement by creating predictions that predict customer behavior, such as web clicks, conversions, or the probability of customer discontinuing a service (churn). By adding predictions to your next-best-action strategy framework, you can determine the next best actions for your customers.

As part of the next-best-action strategy framework, Pega Customer Decision Hub includes several default predictions that cover the most common uses cases. If you want to extend your next-best-action strategy framework, you can create additional predictions in Prediction Studio, and add them to extension strategies. Prediction Studio provides several templates with preconfigured outcomes to help you create predictions for customer engagement:

Prediction templates for customer engagement
The Create a prediction window lists such outcomes as Accept, Click, and Conversion.

Depending on your use case, you can predict one or two outcomes:

1. Single-stage prediction
A single-stage prediction predicts a single outcome, for example, how likely a customer is to click a banner or accept an offer, to convert, or to stop using your service (churn).
2. Multistage predictions
A multistage prediction predicts two outcomes that occur in a sequence, for example, a web banner click followed by a conversion. A multistage prediction requires access to conversion data and long-term feedback. This type of prediction can be useful when groups of customers who click or accept an offer and those who subsequently convert have distinct characteristics.

Choosing between single-stage and multistage predictions

To decide whether a single-stage or a multistage prediction is the better choice for you, your team should answer the following questions regarding your customers, data, and storage:
Data scientistAre the customers who click and the customers who convert similar or different?Conduct a data science analysis to compare the profiles of both groups:
  • If the profiles are similar (except for volumes), you can predict conversions with a single-stage prediction. Clicks are often a good proxy of conversions, so you can use a click model. Alternatively, if you have access to conversion data, you can use a conversion model.
  • If the profiles are significantly different, consider using a two-stage prediction to assess which customers out of those who are likely to click an ad are also likely to convert.
System architectDo you have access to conversion data?You might have to retrieve conversion data from another system, for example, a fulfillment system. This might pose technical challenges, such as a longer waiting period.
Can you retain impression data until customers convert? In systems that process large volumes of data, it might not be feasible to keep the impression data long enough for customers to convert.

To create a prediction, you answer several questions about what you want to predict. Based on your answers, Prediction Studio creates a self-learning adaptive model that is the basis of the prediction. You can replace this model with a different one, for example, a predictive model that you developed in a third-party machine learning platform. You can then include the prediction in your decision strategy.

For example, you can create a prediction that calculates the probability of customer churn, and then add the prediction to a next-best-action strategy. The next-best-action strategy arbitrates several actions to retain the customer, and then selects the one that the customer is most likely to accept.

Pega Customer Decision Hub
  1. In the navigation pane of Prediction Studio, click Predictions.
  2. In the header of the Predictions work area, click New.
  3. In the Create a prediction window, select Customer Decision Hub, and then click Next.
  4. In the Prediction name field, enter a name for the prediction.
  5. Specify what you want to predict.
    For example:

    To predict whether a customer is likely to accept an offer, select the following settings:

    1. In the Outcome list, select Accept.
    2. In the Subject list, select Customer.

    To predict whether a customer who is likely to click a web banner is also likely to accept the corresponding offer and convert, select the following settings:

    1. In the Outcome list, select Click + Conversion.
    2. In the Subject list, select Account.
      Note: A multistage prediction requires long-term feedback. You can define how long to wait for the customer to respond to your offer by configuring the response time-out in the prediction settings.
    Creating a multistage prediction
    The click and conversion template is selected as the prediction outcome

    To predict an outcome that does not match any of the available templates, select the following settings:

    1. In the Outcome list, select Custom.
    2. In the Outcome name field, enter a name for the outcome, for example, Upsell sensitivity.
    3. In the Subject list, select the subject of the prediction.
  6. If you work with branches and want to save the prediction to a specific branch, select Save to branch, and then select a rule set and a branch.
  7. Review your settings, and perform one of the following actions:
    • If you selected Churn as the outcome that you want to predict, click Create, and then go to step 14.
    • If you selected other outcomes, start the Prediction wizard by clicking Start wizard.
  8. In the Select data wizard step, choose one of the following options:
    • To create a prediction without historical data, select I do not have historical data.

      In most cases, historical data is not available or necessary. You can continue without it.

    • To create a prediction with historical data, select I have historical data, and then select the data set that contains your historical data.

      Historical data informs the prediction so that it performs better right from the start. To use this option, you need to import your historical data into a data set. Ensure that the data set contains a pyOutcome property for the outcome of each customer interaction, as well as properties that describe the context of each interaction, for example, pyIssue, pyGroup, pyName, pyDirection, and pyChannel. For more information, see Importing data into a data set.

  9. Click Next.
  10. In the Prediction configuration wizard step, review the response labels for the prediction, and then click Next.
  11. In the Select predictors wizard step, select the fields that you want to use as input for the prediction.
    To increase the accuracy of your prediction, select a wide range of fields to use as predictors. Do not include fields that are not suitable as predictors, for example, the Identifier and Date Time fields. For more information, see Best practices for choosing predictors.
  12. Confirm your choice of predictors by clicking Next.
  13. In the Review prediction wizard step, review the prediction settings, and then complete the model creation process by clicking Create.
  14. To change the prediction properties at this stage, in the Prediction window, click Configure, and then make your changes.
  15. Click Save.
    Result: The prediction is now available in the Predictions workspace. Depending on the settings that you selected, Prediction Studio created an adaptive model or a scorecard as the basis of the prediction.
  16. If you are using parameterized predictors, add them to the adaptive model that is the basis of the prediction.
    For more information, see Adding parameterized predictors.
    Tip: To open the model that is the basis of the prediction, in the Prediction workspace, on the Models tab, click the name of the corresponding model.
  17. Optional: To customize your prediction, for example, by changing the control group settings or by configuring the response time-out, see Customizing predictions.
  18. Optional: To replace the model that is the basis of the prediction with a different model, see Replacing models in predictions with MLOps.
What to do next: Include the prediction in an extension strategy within the next-best-action strategy framework, for example, AIModelsExt or ActionLevelPropensityExt. For more information, see NBA Strategy framework extension points.

Have a question? Get answers now.

Visit the Support Center to ask questions, engage in discussions, share ideas, and help others.

Did you find this content helpful?

Want to help us improve this content?

We'd prefer it if you saw us at our best. is not optimized for Internet Explorer. For the optimal experience, please use:

Close Deprecation Notice
Contact us