RAG for Excel: why spreadsheets are the hardest easy format
Excel cell text extracts cleanly for RAG, but a grid of bare numbers makes poor retrieval material - spreadsheets need more thought about what is worth converting than any other format.
Extraction is easy; usefulness is not
An .xlsx is XML like the other Office formats, so getting the cell contents out is straightforward. The difficulty is that a row of numbers embedded as a passage retrieves badly: '4120 2019 EMEA 0.83' matches almost nothing a person would ask.
Headers are what make a cell meaningful
The number 412000 means nothing; 'EMEA revenue, Q3 2026: 412000' is retrievable. A converter that carries column and row headers into each passage produces dramatically better results than one that dumps the grid, and it is worth checking which yours does.
Consider not converting the numbers at all
If the question you want answered is 'what was EMEA revenue in Q3', a SQL query or a spreadsheet formula answers it exactly, and a vector search answers it approximately. RAG earns its place on the prose around the data - the notes tab, the methodology sheet, the commentary - more than on the figures.
Formulas, not results, are sometimes the content
Where a workbook encodes business logic, the formulas are the documentation. Most converters extract the computed value rather than the expression, so if the logic matters, export it deliberately rather than expecting it to appear.