LLM-Powered Product Supply Chain Lifecycle Decision Support Co-Pilot

Sammanfattning

Supply chain lifecycle decisions involve complex trade-offs between cost, capacity, risk, and timing, particularly during product lifecycle transitions such as ramp-up, maturity, and phase-out. While data-driven decision support systems have long been used to assist such decisions, recent advances in artificial intelligence, especially Large Language Models (LLMs), raise important questions regarding correctness, explainability, and governance in decision-critical industrial settings. This thesis investigates how AI and data science techniques can be integrated into supply chain decision support without delegating decision authority to probabilistic models. A design-science research approach is adopted to design, implement, and evaluate a lifecycle-aware supply chain decision support co-pilot. The proposed system combines deterministic analytical decision engines, machine learning-based risk and pattern insight generation, and a constrained LLM used exclusively for natural-language interaction and explanation. A proof-of-concept implementation demonstrates how lifecycle-aware scenarios can be explored through conversational interaction while preserving deterministic, auditable decision outputs. The system is evaluated using representative industrial scenarios and a structured human expert study, focusing on correctness, explainability, scenario exploration capability, and AI governance. In the comparative evaluation with a generic enterprise copilot, the lifecycle-aware co-pilot is consistently preferred and achieves higher ratings on perceived correctness, faithfulness, completeness, and clarity. The results show that LLMs can enhance usability and decision understanding when strictly constrained, while deterministic models retain full decision authority. This work contributes a safety-first architectural approach and an accompanying evaluation framework for AI-enabled decision support systems, and provides practical insights into responsible AI adoption in industrial supply chain management.