A question regarding NBA success rate (% positive from total responses) vs respective model's performance (AUC). Could very high or low success rate (e.g. 95% customers reply positive/ only 5% reply positive) also negatively impact Adaptive performance? Meaning if majority of customers respond one way then basically all the responses will be "inside the curve" meaning AUC will be high? And the when looking into predictor each bin might have superhigh/superlow propensity which actually is not good. Should the strategy be to have at first as versatile response base as possible in order for Pega to better predict which customers are positive and which negative? Or do You think the underlying problem is still rather lack of good predictor based on what to distinguish positive from negative customers?
This issue occurred now with introduction of new container in digital channel which somehow collects almost only positive responses (although negative response button is also there). This has caused sudden increase in Adaptive scores while Adaptive performance is very low.
***Edited by Moderator Marissa to update Platform Capability tags****