We are in a situation where, for various reasons, our client is considering using one value for CustomerID in MVP, and at some point in the future, may cut over to use a different identifier. I would like to understand what the impact would be to adaptive learning if a customer's CustomerID was to change.
Theoretically, if I have one value for CustomerID today, and tomorrow it is some other value, but that is the only thing that changes about my data, then wouldn't a mature model calculate the same propensities for me today that they would have yesterday?
Correct. CustomerID itself is probably not a predictor, but customer attributes and behavioral data would be, and these don't change.
Perhaps needless to say but for customers in process, there would be an impact. If you make a decision for a customer with ID "X" and try to set a response for that same customer that is suddenly called "Y", obviously that won't work.
To add further details, if you are using delayed-learning approach to train the adaptive models, all Inflight requests (Propositions waiting for the feedback, in-other words CaptureResponse) will be impacted. Please adjust the delayed learning TTL settings if needed to minimize the impact for ADM learning.
Nanjundan Chinnasamy | PEGA Lead Decisioning Architect