Bridging ai-driven risk detection and discrete-event simulation for quantitative supply chain risk evaluation

Abstract

Globally dispersed, project-driven supply chains face increasing disruption risks, while the growing volume of external information makes manual monitoring impractical. Large language models (LLMs) can support the automated detection of disruption-relevant signals in external text sources, but these signals do not directly provide quantitative insight into their potential impact on a specific supply chain configuration. This paper proposes an automated end-to-end methodology that combines LLM-based risk detection with discrete-event simulation (DES) for initial quantitative risk evaluation. Unstructured textual disruption signals are translated into structured disruption vectors and executable SimPy-based scenarios. Through ensemble-based classification, schema-constrained extraction, deterministic validation, and Monte Carlo experimentation, baseline and disruption scenarios are generated and compared automatically. A case study on Red Sea piracy disruptions demonstrates how validated disruption signals can be transformed into measurable lead-time distributions and milestone risk indicators, revealing substantial delay escalation under rerouting conditions. The contribution of the paper lies in demonstrating the feasibility of a fully automated pipeline that converts external disruption signals into initial quantitative risk indications for further expert assessment in complex, project-oriented supply chains.