ADM Model instance created with Action Name as JSON and Treatment blank.

General
, , , ,
1

Hi CDH Product Team,

We are using Pega Infinity 24.1.2, We are using OOTB model rules i.e Web_Click_Through_Rate and Mobile_Click_Through_Rate. At runtime the Action and Treatment are mapped properly but for some model instance we could see Action column has JSON content and Treatment Column is blank.

I am looking forward to understand :

  1. Why it is happening, even if the Model Context is defined properly ?
  2. Why it is happening only for some of the model instance ?
  3. Given that 2 model instance created for same Action + Treatment combination, Which one is gonna be used for Arbitration at run time ?

Form Verizon hyper-personalization stand point it is very critical to have ADM model instance created and learning for treatments with multiple variants. So if this is a defect which need to be fixed, please let me know the next steps.

Note : I can not attach any screenshot.

–

Thank you

Prasanna Kumar

2

@PK2013137 the reason you cannot attach screenshots is likely because you did not log into your Pega Account.

Make sure you see 'Hello ’ on the top right menu of the portal.

:warning: This is a GenAI-powered tool. All generated answers require validation against the provided references.

Understanding the Issue

This behavior appears to be related to a format change introduced in Pega Platform 24.1. Previously, in earlier versions of Pega Platform, when treatment was one of the model identifiers that defined the model context, the action and treatment details were combined in reports and displayed in JSON format. In 24.1, Pega updated the format to display treatment information in a separate column to improve readability and to make it easier to filter and sort model data.

The JSON content you’re seeing in the Action column (with blank Treatment column) is likely showing the old format still being used for some model instances, while others have been converted to the new format. This mixed behavior suggests an incomplete conversion of the model instances to the new reporting format.

Why This Is Happening

  1. Why it’s happening even with properly defined Model Context:
    The issue is likely not related to your Model Context definition but rather to the system’s handling of the new reporting format. Some model instances may not have been properly converted to display treatment information in the separate Treatment column.
  2. Why it’s happening only for some model instances:
    This could be occurring because the system conversion was only partially completed. New model instances might be created with the proper format, while existing ones or those created under specific conditions might still use the old JSON format.
  3. Arbitration with multiple model instances:
    When there are multiple model instances for the same Action + Treatment combination, the system evaluates the propensity scores from each relevant model instance during arbitration. However, if some instances are using the old format and some the new, it could potentially affect how these models are identified during arbitration, which might be why you’re concerned about this issue for hyper-personalization scenarios.

Next Steps

Since this appears to be related to a formatting change in Pega 24.1.2 and is affecting your critical Verizon hyper-personalization work, I recommend:

  1. Run the conversion activity: Starting with Pega Platform 24.1, there’s a conversion activity available to update the report format for existing model instances. This may resolve your issue by ensuring all instances use the new format consistently.

To do this:

    • Navigate to Prediction Studio and click on Models
    • Click on any Adaptive Model name
    • Look for an option to convert the report format (note that this can be run from any Adaptive Model details page)
  2. If the conversion doesn’t resolve the issue: Create a support case with Pega Support to report this as a potential defect.
  3. Be sure to to provide the INC id here after you log it via the MSP. IN the MSP ticket details include:
    • Screenshots showing the inconsistent behavior
    • Details about the specific model instances affected
    • Information about any impact you’ve observed on arbitration
    • The urgency related to the Verizon hyper-personalization project
  4. Temporary workaround: Monitor both columns (Action and Treatment) during arbitration testing to ensure proper model selection. While not ideal, understanding how the system is interpreting these different formats will help ensure proper functioning of your hyper-personalization implementation.

[References: