Currently in NBA designer we have noticed there is customer hashing and setting the propensity as below for action and treatment. We are not sure why this process introduced and also not finding any extension to enable or disable this functionality.
- Hashing the customerbankid as below for action or each channel treatment
@hashToProportion(.OriginalContactID+@month(@today())+@year(@today()))
- Verify the hash value and set the ModelControlGroup value to “Control”
@if(.pySubjectHash>=0.0&&.pySubjectHash<=0.02,“Control”,“Test”)
- Verify current customer fall under model group then set the random propensity
@If(.ModelControlGroup==“Control”,@random(0,1.0),.pyPropensity)
Can some one provide purpose of this implementation and how this related to control group.
Thanks,
Deva
@DevarajPrathapaneni Thank you for your question.
This logic is used internally by the NBA Designer Strategy Framework when using Adaptive Modelling to calculate propensity for NBAs. It implements a 2% control group using the hashing method you highlight. Subjects captured in the control group are assigned a random propensity for the NBA which then undergoes Thompson Sampling. This provides the Adaptive Models with the continued opportunity to learn increasing the chances of identifying new opportunities for well-placed engagement and reacting to new trends.
This approach has been in place for many Pega Infinity 8 versions up into 8.8 where this logic is intended to work with Impact Analyzer to provide uplift insight for Adaptive Modelling.
See here for more information about how we use Control Groups and Thompson Sampling in the framework with Adaptive Models.
8.8:
8.7: