Should I re‑run prediction monthly or use single snapshot at MOB3 to avoid customer group movement in N2B banking model?

External question — Cross Validated Stack Exchange Author: Alice wong Original post: https://stats.stackexchange.com/questions/676940 License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/ Adaptation: HTML converted to plain text; contact email addresses removed. I am working on a binary classification model for New‑to‑Bank corporate customers in a bank. Business context (brief): We want to predict whether zero‑product N2B customers will convert to use payment‑cash‑management products in future months. Observation window is customer MOB1‑MOB3 onboarding behaviour; prediction target is conversion within MOB4‑MOB12. My technical question: Two modelling approaches are considered: Re‑train / re‑score every calendar month for each MOB bucket. This allows updating customer potential every month, but causes customer group movement : same customer can flip between high‑potential / low‑potential segments month‑on‑month. Run prediction only once at MOB3 , using MOB1‑MOB3 features. The customer segment label is fixed for the whole MOB4‑MOB12 period; no monthly re‑scoring, so no group flipping. What are statistical pitfalls, pros and cons for these two approaches? Are there standard practices in banking analytics to handle customer‑group‑movement when applying predictive model outputs for downstream marketing segmentation? I am not asking for full project design; I only want to understand the statistical / modelling trade‑offs between these two scoring cadences.