I’m doing some customer research around a workflow I keep seeing:
People receive data in PDFs, but the real work happens in a spreadsheet.
Not necessarily a clean “PDF to Excel” conversion — more like:
I’m trying to understand how people handle this today.
A few questions:
When you receive a PDF and need the data in Excel or Google Sheets, what do you actually do right now?
Do you usually need a brand-new spreadsheet, or do you need the data placed into an existing template?
Which part is most annoying?
I’m especially interested in workflows around invoices, bank statements, purchase orders, bookkeeping, operations, or any recurring back-office spreadsheet work.
Not trying to pitch hard here — I’m trying to understand the existing workflow before deciding what to optimize for.
The template-matching part feels like the real workflow, not the PDF conversion. In Tokens Forge's AI Researcher, the useful output is not just a finished report; it is the source trail, uncertainty, and enough structure that the user can audit what happened. For PDF-to-spreadsheet, I would want the AI to return three layers: extracted fields with page/source references, a normalized table mapped to my existing columns, and a review view that flags low-confidence cells before export. Also worth showing the token/run cost for large PDFs, because recurring back-office jobs can quietly become expensive.
This is a very useful breakdown. I agree that template matching is closer to the actual job than PDF conversion itself.
The three-layer model makes a lot of sense: source-linked extraction, mapping into the user’s existing columns, and a review layer that highlights uncertain cells before export. The review layer is especially important because the goal should not be to pretend the AI is always correct, but to make verification faster and more focused.
I also like the idea of showing processing cost upfront for larger PDFs. ParseToSheet currently uses a credit-based model, so making the expected cost clearer before a run would help users judge whether a recurring workflow is economical.
Thanks — source references and confidence-based review are now two areas I want to explore more seriously.