Skip to main content


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

Creating adaptive models at different hierarchy levels

Updated on July 5, 2022

Adaptive models are self-learning models that you can use in strategies to improve predictions about changing customer interests and needs.

In a standard configuration of an adaptive model rule, the rule context reflects the proposition hierarchy so that a unique adaptive model is generated for every combination of Business issue, Group, Proposition, Direction, and Channel values. In many situations, you might need to add or remove particular model identifiers from the adaptive model rule to increase or decrease model granularity.

This tutorial explains how to create feature-based adaptive models for two groups of consumer financial products instead of having six adaptive models for each type of loan and credit card. Adaptive models for particular products in a group are so similar that is preferable to have one adaptive model at the group level.

Strategy designers can change the standard setup of the adaptive model rule and customize model identifiers in a decisioning application. Model identifiers that are customized in the adaptive model rule are called flexible partitions and can be used in the following situations:

  • You introduce new offers frequently. If you create offer-level models, you experience the cold-start problem. Because of the lack of initial evidence for adaptive models, the models need some time before they can collect customer responses.
  • Your propositions are not offered very often (for valid reasons), which results in slowly learning models.​ At the same time, you might have similar propositions with many more responses and you would like to consolidate all the responses for one model.
  • You have propositions that are similar to each other so they will be perceived by customers in a very similar way.
  • The proposition hierarchy that you use does not match naturally with the organization of products or offers.
  • You want to create more accurate models and have enough response data, so you introduce another level in the model partitioning. For example, in marketing strategies you can add model identifiers such as Treatment or Call Reason.
  • You need to disregard the standard hierarchy in favor of feature-based models. For example, you need adaptive models for product attributes such as brand, price range, or color.
  • You need to limit the number of models because there are too many propositions to handle with the Adaptive Decision Manager cluster, or you want to monitor and review models that have very similar behavioral patterns.​
Before you begin:

Create a delayed learning setup with data flows and strategies. For more information, see Delayed learning of adaptive models.

This tutorial uses the following strategies and propositions that are referenced in the delayed learning setup:

MakeDecision strategy
The strategy consists of the following components: proposition data, adaptive model, prioritize, and results.

In the Proposition Data component, you import all the propositions from the Consumer Financial Products business issue. In the Adaptive Model component, you reference the Banking Products adaptive model. In the Prioritize component, you configure the following options:

Prioritize component
The component references the py propensity expression. The output setting is top three.
Response strategy
The strategy contains the following components: external input, set outcomes, results.
pyNameGroup
Platinum CardCredit Cards
Gold CardCredit Cards
Silver CardCredit Cards
Cash Advance LoanPersonal Loans
Home Equity LoanPersonal Loans
Vehicle LoanPersonal Loans

To create adaptive models for two groups of consumer financial products, customize flexible partitions on the Context tab of the adaptive model rule.

Note: In this tutorial, you create adaptive models at the group level for two groups with six consumer financial products in total. However, you can also build adaptive models at the group level when you have hundreds or thousands of products (propositions) in different groups.
  1. In the header of Dev Studio, click CreateDecisionAdaptive Model.
    For example: Consumer Financial Products
  2. Click the Context tab and delete the .pyName model identifier.
    Caution: ​Be careful when you modify flexible partitions. Increasing the number of model identifiers might lead to a large number of adaptive models. You can get as many models as all the possible combinations of distinct values for each model identifier.
  3. Click the Outcomes tab and specify positive and negative outcomes for the model.
  4. Click Save and close the rule instance.
  5. Run the MakeDecision strategy to create models for financial products.
  6. Optional: To view adaptive models in the Model Management landing page, click ConfigureDecisioningModel Management.
    Two adaptive models for groups of consumer financial products
    One model belongs to the card products group, the second model belongs to home loans.
  7. Optional: View an overview of adaptive models by using reports.
What to do next: After you modify the standard hierarchy and add flexible partitions in the adaptive model rule instance, do an experimental analysis of the models. Verify that the extra dimensions work as intended and that the models get enough responses and collect enough evidence.

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.

Pega.com is not optimized for Internet Explorer. For the optimal experience, please use:

Close Deprecation Notice
Contact us