In the era of Industry 4.0, digital twin (DT) has gained substantial attention across industries for its capability to create dynamic digital representations of physical systems. A DT facilitates continuous data exchange between a physical asset and its virtual counterpart, enabling real-time monitoring and predictive analysis. A critical precursor to developing a full DT is digital shadow (DS), which digitally mirrors the behavior of a physical system using simulation tools. The effectiveness of a DS largely depends on the accuracy and fidelity of the input data. This paper proposes a step-by-step methodology for developing a DS using Discrete event simulation (DES) in FlexSim. The approach begins with a product architecture analysis to identify key process variables, enabling effective product family classification. This is followed by the selection of suitable DES software, with a focus on accommodating high product variability typical in complex manufacturing systems. A dual-mode data acquisition strategy, combining manual time studies with sensor-generated data, is then applied to capture both observable and high-resolution operational data. A comparative case study is conducted using two DS models, one based on manually collected data and the other on high-fidelity sensor data, within a high-mix, high-volume manufacturing setting. Results demonstrate that DS models built with high-fidelity data significantly enhance simulation precision. The second model successfully identified a production line imbalance and provided actionable insights through a micro study of manual activities that were executed in the assembly process. Furthermore, the proposed future state, modeled through the DS, showed a potential 20 % increase in production output, contributing to operational efficiency. This study aims to underscore the value of DS as a reliable and scalable tool for analyzing complex systems, supporting informed decision-making, and serving as a critical step toward the implementation of a complete DT.