This study enhances CLV prediction in fashion e-commerce through machine learning models trained on 4.2 million transactions from a leading European retailer. Building on [1], gradient boosting achieves 89% precision in 12-month CLV forecasts, outperforming Pareto/NBD baselines by 18% RMSE reduction. The framework integrates real-time browsing patterns and returns reasons (30% order return rate), demonstrating that sizing-related returns correlate with 73% repurchase intent when addressed proactively. Dynamic micro-segmentation strategies, validated by [2], yield 31% reduction in stockouts through CLV- driven inventory allocation. The research addresses post-2020 challenges: non- contractual relationships (40% single-purchase customers), pandemic-induced demand spikes, and ethical implications of value-based segmentation. Operational insights include RNN-based churn prediction models that reduce customer acquisition costs by 22% through targeted retention campaigns. The paper concludes with MLOps protocols for weekly model retraining to counter fast-fashion trend volatility (48-h feature drift cycles).