ML-Driven Digital Twin for Supply Chain and Cash Flow Optimization Under Demand Uncertainty

Abstract

Modern retail supply chains operate under significant uncertainty due to fluctuating demand, variable lead times, and dynamic cost structures. Traditional inventory management approaches rely on static reorder policies based on historical averages, which fail to adapt to real-time demand patterns and often lead to suboptimal outcomes such as stockouts or excess inventory.This paper presents an ML-driven digital twin framework for joint optimization of inventory management and cash flow in a retail supply chain setting. The proposed system integrates demand forecasting using machine learning models, including Linear Regression and Long Short-Term Memory (LSTM) networks, where comparative analysis shows that simpler models outperform deep learning approaches for the given dataset, with a discrete-event simulation engine that models day-level inventory dynamics, replenishment decisions, and financial transactions.The framework evaluates multiple inventory control policies, including a fixed baseline policy, an adaptive policy driven by ML predictions, and an adaptive policy augmented with statistically derived safety stock. Additionally, Monte Carlo simulation is employed to capture demand uncertainty and quantify risk in terms of stockout rates and cash flow variability.Experimental results on a real-world retail dataset demonstrate that the adaptive policy consistently outperforms the baseline approach by improving service levels and stabilizing cash flow, while the inclusion of safety stock significantly reduces stockout risk at the expense of higher holding costs. The proposed digital twin framework provides a robust decision-support tool for supply chain optimization under uncertainty.