We’re using a rest-service and a real-time container to deliver a nba request.

The click through behavior of the container is configured as capture click through only

The container captures impression by channel. So the container does not write records in Interaction history.

In the strategy we configured an adaptive model. When starting the marketing campaign with this container a dataflow (eg ContainerDFP1234)   is generated.

This dataflow has the decisions strategy component configured with “Make decision and store data for later response capture”

When we send a nba request message an interactionid is generated in the container response and models are generated on the adaptive models landing page.

For processing the response (feedback) we use another rest-service.  

The message for this rest-service contains the interaction id from the previous container response.

The rest-services stores the response in Interaction history using the dataflow Df_CaptureInteractionHistory

This dataflow calls the dataflow DF_ProcessResponse (configured with pxAdaptiveAnalytics)

and feeds the response to the pxAdaptiveAnalytics dataset. By doing this, the generated adaptive models are updated.

So the model should learn from the responses.

This scenario differs in a number of ways from the examples used in the adaptive modeling courses on Pega Academy.

We don’t have a dataflow with a response strategy configured when we process the response.

In the course the pxAdaptiveAnalytics dataset is added to this dataflow (instead of the ProcessResponse dataflow).

According to the course the response strategy in the flow should be configured with “Capture Response from previous decision with interarction id”

So we wandered if our approach is a valid way to use adaptive models. Or should we use it as described in the courses on Pega Academy ?