Various organizations’ stakeholders often base their decisions on insights extracted from financial documents such as invoices. Most of the time, these are non-technical stakeholders, thus requiring a translation into a structured database query. This paper explores two approaches that leverage the current capabilities of LLMs (Large Language Models) and facilitate natural language querying applied on factual data. We compared the established RAG retrieval method with its extension, GraphRAG, as both of these methods empower decision makers to quickly address natural language queries at their convenience to obtain actionable insights from invoices, reducing reliance on specialized personnel for database interrogation.