Zalando's ZEOS replenishment optimization system has improved gross merchandise value (GMV) by 22.1% by integrating probabilistic demand forecasting with a risk-aware discrete event simulation. This system addresses the Inventory Paradox in e-commerce, where too much stock ties up capital, while too little leads to stock-outs. Their recent publication in *Nature Scientific Reports* outlines how they moved beyond simple forecasts to create a complex simulation-driven engine that optimizes inventory under uncertainty, using an extended (R, s, Q) policy. The ZEOS tool relies on three main components: a forecaster using LightGBM that models full probability distributions of demand, an engine that adjusts reorder policies dynamically throughout a product's lifecycle, and an optimizer that simulates thousands of potential futures for each policy. The system tracks various costs, including storage and lost sales, minimizing not just average costs but also the risk of extreme demand spikes. Over a year-long backtest involving around 2 million articles from 800 merchants, the engine consistently outperformed human decisions, achieving significant improvements in availability and demand fill rates. Comparative analyses against traditional replenishment methods highlight the effectiveness of Zalando's approach. The extended (R, s, Q) policy outperformed conventional methods like the Myopic Newsvendor and Periodic base-stock strategies, with GMV uplifts of 22.11% compared to just 5.07% for the Myopic Newsvendor. The results indicate that around 70-80% of merchants in the study benefited from this probabilistic method, showcasing its capacity to optimize inventory management across diverse product assortments.