While customer segmentation is fundamental to personalized marketing, conventional approaches offer only a static view of the customer base. This creates a critical disconnect between historical analysis and the need for immediate action, as segments quickly become obsolete in the fast-paced e-commerce environment. To address this, our paper introduces an end-to-end framework that transforms unsupervised clustering results—derived from RFM metrics and a hybrid K-selection method combining data-driven analysis with business heuristics—into a high-performance supervised classification system. Validated on a large-scale retail dataset, this framework, powered by a top-performing XGBoost model, achieves 99.0% accuracy in predicting dynamic customer segments. The framework’s core innovation lies in its self-adaptive mechanism, which enables instant classification upon any transaction and incorporates a scheduled retraining strategy to combat concept drift. Ultimately, this work provides a practical and scalable blueprint for businesses to finally bridge the gap between static customer understanding and dynamic, actionable marketing intelligence.