Using Generative Artificial Intelligence as a Decision Support Tool in Purchasing Processes: Comparison of ChatGPT, CoPilot, and Gemini Tools

Abstract

Accurate prioritization of purchase requests in enterprises is critical for ensuring business continuity and effective resource management. Throughout the day, requests generated by different departments are usually ranked subjectively by the purchasing unit, which may cause some urgent requests to be deprioritized. Managing the process under human control leads to time loss and inaccurate prioritization. This study integrated three generative artificial intelligence tools—ChatGPT-4.5, Microsoft CoPilot, and Google Gemini—into a manufacturing company’s ERP system via an API. A total of 100 purchase requests were classified first into three categories (“Urgent,” “Normal,” and “Not Urgent”) and then into two categories (“Urgent” and “Normal”). The results produced by the AI models were compared with the classifications made by the purchasing staff and evaluated using accuracy, Cohen’s Kappa, precision, recall, and F1-score metrics. In addition, the correct response performance of generative artificial intelligence tools was analyzed using the Pearson Chi-square test; the results revealed a significant interdependence among the tools, with Copilot and Gemini showing an exceptionally high consistency across both triple and binary classifications. The findings revealed that all three models performed well in the binary classification, with CoPilot achieving higher accuracy than the others. The study demonstrates that generative AI tools can be practical decision-support systems in purchasing processes, offering significant advantages in preliminary classification, efficiency, and time savings.

Ähnliche Arbeiten