With the expansion of operational data and the growing complexity of business environments—particularly in the retail industry—accurate demand forecasting has become a key requirement for effective decision-making and efficient supply chain management. The nonlinear nature, high volatility, and noise inherent in sales data often lead to poor and unreliable performance of traditional forecasting methods. This challenge is further amplified in Iran due to the high level of uncertainty arising from economic fluctuations, shifts in trade policies, logistical constraints, and the impacts of sanctions. In response to these needs, the present study introduces the GODFIS algorithm, a Global Optimization-based Dynamic Fuzzy Inference System, designed to provide accurate, robust, and adaptive forecasts in highly volatile environments. The architecture of GODFIS—through the integration of online learning, dynamic clustering, recursive updating of consequent parameters, and a noise-filtering mechanism—enables precise modeling of complex relationships without the need for full retraining. To evaluate its performance, the proposed algorithm is tested on two categories of data: (1) benchmark datasets with non-stationary behaviors, and (2) real-world data from an Iranian retailer. The results indicate that GODFIS delivers substantial improvements in accuracy, stability, and adaptation speed compared to classical methods, machine learning models, and evolving systems. Overall, the findings suggest that GODFIS, by offering interpretability, scalability, and computational efficiency, can serve as a dynamic forecasting framework for retail operational environments—especially under Iran’s unstable conditions—and play a significant role in optimizing inventory decisions and managing uncertainty.