A data-driven framework for enterprise management and using system dynamics, machine learning, and computational modeling

Data-driven enterprise management in the artificial intelligence era demands modeling paradigms that are capable of handling dynamic enterprise interactions and nonlinear data relationships in complex enterprise settings. Although the hybrid system dynamics–machine learning (SD–ML) modeling paradigm in the existing work has shown excellent predictive power in energy systems. In addition, it still has limitations in terms of domain dependency, the lack of enterprise-level data, the poor integration of actual enterprise transaction data, and the absence of enterprise policy scenario analysis. On the other hand, the recent development of enterprise-level multi-regional input–output (EMRIO) databases has made available global supply chain data at the firm level with high granularity, but without dynamic decision modeling. To address the existing limitations, this paper introduces a new data-driven enterprise management framework that combines the hybrid SD–ML modeling paradigm with the EMRIO enterprise-level data. The proposed workflow initially starts with the construction of enterprise causal loop and stock-and-flow diagrams to model resources, production, demand, and performance feedback. Next, EMRIO-based enterprise variables and transaction signs are preprocessed to map them to dynamic state variables. In addition, a light gradient boosting machine model is developed to identify nonlinear relationships and predict enterprise performance indicators. Finally, machine learning outputs are used to calibrate and update the system dynamics simulation for scenario analysis and policy optimization. The experimental results on the EMRIO enterprise dataset show that the hybrid SD–ML model outperforms benchmark models in terms of predictive accuracy, goodness of fit, and forecasting error for enterprise performance indicators.