What is it about?
Choosing the right supplier is one of the most complex decisions companies face, requiring evaluation across cost, quality, delivery, sustainability, and more. This study presents a new framework that combines traditional scoring methods with a language model (DistilGPT-2) to not only rank suppliers numerically but also generate plain-language explanations of why a supplier is strong or weak. Tested on 362 Vietnamese textile companies, the system agreed with expert human evaluators 78.3% of the time and produced far more understandable outputs than conventional approaches, making it practical even for small businesses without dedicated data science teams.
Featured Image
Photo by Timelab on Unsplash
Why is it important?
Most AI-based supplier tools tell you who to pick but not why leaving procurement teams unable to explain or defend their decisions. This framework closes that gap by turning numbers into readable narratives, achieving an interpretability score of 4.2/5 compared to 1.8–2.8 for traditional methods. It is also the first validated LLM-based supplier selection system benchmarked against real expert judgments, offering a reproducible baseline that other researchers and practitioners can build on. Because it runs on a compact, open model, it is accessible to SMEs that cannot afford large-scale AI infrastructure.
Perspectives
What motivated this work was a straightforward frustration: companies spend enormous effort scoring suppliers, then still struggle to communicate the reasoning to stakeholders. We wanted to show that a lightweight language model — one small enough to run without cloud-scale resources — could bridge that communication gap without sacrificing accuracy. Seeing 87% of procurement experts call the outputs helpful was genuinely encouraging. We hope this sparks wider exploration of interpretable AI in supply chain management, especially for organizations that have been left behind by tools designed only for large enterprises.
Demiral Akbar
OSTİM Technicail University
Read the Original
This page is a summary of: Large Language Model-Enhanced Multicriteria Decision-Making for Supplier Selection: A Validated Framework with Statistical Analysis, Smart and Sustainable Manufacturing Systems, March 2026, ASTM International,
DOI: 10.1520/ssms20250044.
You can read the full text:
Contributors
The following have contributed to this page
