An Empirical Study on the Feasibility Analysis of On-Premise RAG for AI Diffusion in Manufacturing

This study empirically validates the feasibility of On-Premise Retrieval-Augmented Generation (RAG) as a techno-managerial alternative for SMEs facing security and cost barriers in AI adoption. Applying the TOE (Technology-Organization-Environment) framework, we conducted a comparative analysis between a Base Model (Zero-shot), GPT RAG, and an open-source-based On-Premise RAG. Results indicated that the On-Premise RAG not only matches commercial models in quantitative performance but also superiorly fulfills qualitative criteria such as usefulness and relevance by utilizing a systematic knowledge-base approach. While exhibiting inherent latency, the system optimizes organizational efficiency by eliminating recurring token costs and ensures environmental security by fundamentally preventing external data leakage. Ultimately, On-Premise RAG provides a reliable, cost-effective solution for maintaining data sovereignty in specialized manufacturing domains.