A Stakeholder-aware Decision Support System for Make-to-Order Production Planning Using Retrieval-Augmented Generative Artificial Intelligence

Journal of Systems Thinking in Practice

دوره 5، شماره 3 - شماره پیاپی 18، پاییز 2026، صفحه 1-25   

نوع مقاله: Research Article

شناسه دیجیتال (DOI): 10.22067/jstinp.2026.98779.1206

نویسندگان

Mozhde Rahjoo^* ؛ Farzad Haghighi Rad؛ Amir-Reza Abtahi

Department of Operations Management and Information Technology, Faculty of Management, University of Kharazmi, Tehran, Iran.

چکیده

Make-to-order (MTO) manufacturing involves production planning decisions under uncertain orders, incomplete operational information, and conflicting stakeholder priorities. To address the need for traceable and defensible cross-functional alignment, this study develops a stakeholder-aware, retrieval-augmented-generation-enabled decision support system (DSS) operationalized through a mixed-methods industrial case study. The system integrates empirical data from 200 orders, 279 downtime events, and 400 quality control, nonconformity, and optical testing records. Rather than relying on probabilistic models for execution calculations, the architecture enforces deterministic release gates to evaluate organisational release readiness prior to optimisation. Retrospective evaluation reveals that the gates classified organisational release status, releasing 170 orders and placing 30 orders on HOLD due strictly to quality control (QC) and optical time-domain reflectometer (OTDR) violations. For the organisationally releasable orders, the engine evaluates two benchmark policies (BASE_FIFO and BASE_EDD) together with three generated scenarios (PROFIT, EFFICIENCY, and CONSENSUS) using the implemented E2 stakeholder-weighted evaluation model. The objective weights were obtained through direct preference elicitation (production efficiency = 0.4688, profit = 0.2997, and value added = 0.2314). The baseline policy (BASE_EDD) benchmark achieved the highest E2 stakeholder-weighted score of 0.5664 and maintained an OTD rate of 90.0% for the evaluated dataset. By restricting the local Large Language Model (Llama-3-8B) strictly to an explainable AI (XAI) interaction layer, the system generates source-grounded justifications without altering deterministic data. The primary academic contribution of this research is the structural separation of organisational release control from multi-objective scenario evaluation, providing a verifiable socio-technical framework for complex make-to-order planning.

کلیدواژه‌ها

Make-to-order manufacturing؛ Systems thinking؛ Decision support system؛ Retrieval augmented generation؛ Generative artificial intelligence؛ Stakeholder Alignment

مراجع