SCSimulator: An Exploratory Visual Analytics Framework for Partner Selection in Supply Chains through LLM-driven Multi-Agent Simulation

TL;DR

SCSimulator introduces an interactive visual analytics framework that harnesses LLM-driven multi-agent systems to support partner selection in supply chains. By combining a four-stage MAS pipeline, an explainable AI layer using SHAP and CoT, and egocentric, timeline-based visualizations, the system enables exploratory scenario analysis with human-in-the-loop control. Formative studies with domain experts yielded seven design requirements, guiding a backend frontend architecture that can map data, simulate adaptive networks, and forecast horizon features. Technical and user evaluations show the approach provides credible, usable sense-making for SC decision-makers, while recognizing limitations in scalability and the need for expert oversight before real-world deployment. The work offers methodological and practical foundations for future AI-assisted, interactive decision support in complex supply networks.

Abstract

Supply chains (SCs), complex networks spanning from raw material acquisition to product delivery, with enterprises as interconnected nodes, play a pivotal role in organizational success. However, optimizing SCs remains challenging, particularly in partner selection, a key bottleneck shaped by competitive and cooperative dynamics. This challenge constitutes a multi-objective dynamic game requiring a synergistic integration of Multi-Criteria Decision-Making and Game Theory. Traditional approaches, grounded in mathematical simplifications and managerial heuristics, fail to capture real-world intricacies and risk introducing subjective biases. Multi-agent simulation offers promise, but prior research has largely relied on fixed, uniform agent logic, limiting practical applicability. Recent advances in LLMs create opportunities to represent complex SC requirements and hybrid game logic. However, challenges persist in modeling dynamic SC relationships, ensuring interpretability, and balancing agent autonomy with expert control. We present SCSimulator, a visual analytics framework that integrates LLM-driven MAS with human-in-the-loop collaboration for SC partner selection. It simulates SC evolution via adaptive network structures and enterprise behaviors, which are visualized via interpretable interfaces. By combining CoT reasoning with XAI techniques, it generates multi-faceted, transparent explanations of decision trade-offs. Users can iteratively adjust simulation settings to explore outcomes aligned with their expectations and strategic priorities. Developed through iterative co-design with SC experts and industry managers, SCSimulator serves as a proof-of-concept, offering methodological contributions and practical insights for future research on SC decision-making and interactive AI-driven analytics. Usage scenarios and a user study demonstrate the system's effectiveness and usability.

Figures (9)

• Figure 1: Framework of SCSiumulator. It integrates an LLM-driven multi-agent system with interactive visual analytics. Users map raw data, configure parameters, run simulations, and iteratively refine agent behaviors through a human-in-the-loop workflow.
• Figure 2: Pipeline of LLM-driven MAS. Each firm is modeled as an autonomous LLM agent that collaboratively plans, queries, requests, and replies to others. The four-stage request–response cycle simulates partner-seeking behaviors and the evolving dynamics of SC networks.
• Figure 3: The SCSimulator simulation interface incorporates five main views, each serving a distinct purpose. The Control Panel aims to support model selection and parameter configuration. The Global View aims to reveal the temporal evolution and relational dynamics of the SC structure. The Focus View aims to reveal how a focal enterprise interacts with its partners, highlighting influence balance and relationship dynamics within SCs. The Adjustment View aims to trace and adjust agents’ behaviors and interactions throughout the simulation. The Simulation Path View aims to visualize simulation progress and enable exploration of alternative paths.
• Figure 4: The Upload Page of SCSimulator. Users upload raw datasets and collaborate with the LLM agent through the Conversation Viewto map data into structured templates. The Data Display Viewpresents real-time mapping results for review and modification before proceeding to the next step.
• Figure 5: The Focus View Design. The overall layout arranges elements chronologically from top to bottom. At each timestamp, suppliers are on the left, the focal company in the center, and customers on the right; each row shows one focal company's SC for comparison. Red dashed lines mark shared suppliers across focal companies. Industry-grouped partners use a berry metaphor: raspberries (warm tones) for upstream suppliers and blackberries (cool tones) for downstream customers; the lower "soil" half shows the focal company's influence on the group, while the upper "berries" half depicts individual partners' influence on the focal company. Colors encode collaboration stages (initiate, maintain, terminate). The focal company uses a rose chart glyph: the central yellow circle encodes performance (larger radius for higher values), and outer arcs represent features like technology, operation, and reputation (radii encode magnitudes). The metaphor overview: the focal company as the sun, with upstream and downstream partners as berry bushes on either side.