Embedding Retrieval-Augmented Generation into SME Customer Communication Work Practices
Embedding Retrieval-Augmented Generation into SME Customer Communication Work Practices

Summary/Abstract: The increasing use of digital messaging platforms such as WhatsApp, Instagram, and Facebook has made real-time customer communication a key activity for Small and Medium Enterprises (SMEs). However, many SMEs struggle to access and apply business knowledge during live interactions, often relying on fragmented information and keyword-based retrieval that does not capture user intent. This study examines how a mobile keyboard–based Retrieval-Augmented Generation (RAG) system supports SME customer response work practices from an Information Systems perspective. The system organizes business knowledge into semantic chunks, represents them as vector embeddings, retrieves relevant information using similarity-based methods, and generates context-aware responses through a Large Language Model. It is implemented as a lightweight keyboard interface embedded directly within messaging applications. The system was evaluated using the RAGAS framework on 37 test queries and compared with a keyword-based baseline. The results show high faithfulness (0.997) and answer correctness (0.881), with an average response time of approximately 5 seconds. A preliminary User Acceptance Testing session with one SME stakeholder suggests that the system may help reduce response effort and support more consistent use of business knowledge in customer communication.

• Details
• Contents