Integrating Generative AI to Drive Efficiency in Spend Intelligence and Negotiation Strategy

This paper explores the integration of generative artificial intelligence (AI) technology into the procurement operations of a global healthcare company. Driven by a large procurement spend of $35BN with diverse sourcing information and massive amounts of data, the research aims to help our sponsor company develop a real-world proof-of-concept of generative AI that can be successfully implemented in the procurement function. With this objective in mind, we developed a chatbot that democratizes data mining skills to category managers and promotes smarter supplier negotiations. We implemented a retrieval augmented generation (RAG) approach which is less computationally expensive and reduces hallucinations. We used LangChain's text-2-SQL agent on the sponsor’s company relational database architecture. In parallel, we used LangChain's kuzuQAchain agent on a graph knowledge database architecture that we created using a Python library called Kuzu. The model takes the natural language queries and generates either SQL code or Cipher code (depending on the question type), retrieves the relevant information, and returns to the user's natural language with the answer to the prompt. We managed to develop a model that does not return false answers or hallucinations. The final prototype has been presented to the sponsor company and potential users who highlighted the promising benefits that this solution will offer when deployed in terms of efficiency and insight-gathering capabilities.