Question
Merkle Inc
US
Last activity: 14 Sep 2020 6:41 EDT
Adaptive Models
I have done several ADM implementation assessments, and keep seen the same mistake over and over. The outcome behavior options are POSITIVE and NEGATIVE, when it should be POSITIVE and NON-POSITIVES.
The reason for this is that a model predicts the happening (1) or not happening (0) of an event.
On ADM use cases for call centers, this approach was OK as positive=accepted, negative=declined, but on web, email, and other use cases this is no longer valid.
For example, on web, the prediction is an ad click, where the positive behavior is a click, and a negative behavior is an impression minus the clicks.
Are the above statements OK? If not, what is the right approach for use cases where the user will not give a direct 'negative' response?
Pegasystems Inc.
IN
Hi,
I would suggest leveraging the identity matching capabilities explained here
https://community.pega.com/video-library/leveraging-identity-matching
This demonstration will describe how to design strategies that recommend relevant offers by leveraging the identity matching capability.
Hope this helps!
Cheers,
Santhosh
Pegasystems Inc.
NL
Hi Leonardo,
I agree with your statements.
In recent versions of CDH, we have addressed this in the Next-Best-Action Designer framework which contains OOTB adaptive models and applies best practices for those, which may differ per channel. The user then no longer needs to define the possible outcome labels. We also now make use of a response waiting time mechanism, which is used for example to handle the response to an email (if no email has been received within N days, a negative response is captured).
At the same time it's still perfectly OK to use the pattern where web impressions are registered as negatives and a click as positive, it's an accurate approximation and efficient pattern to use,
Regards,
Ivar