PDF to Excel

Tables in PDFs are a notorious pain point. The PDF format has no concept of a table — it just has absolutely-positioned text fragments and lines. To rebuild a table, a converter has to cluster text fragments into rows and columns based on their x/y coordinates, infer table boundaries from line positions, and handle merged cells, spanning rows and rotated headers.

For clean financial statements and data appendices exported from a spreadsheet, this works well. For tables that were photographed, scanned with skew, or built with custom drawing commands (vertical text, diagonal headers), the output is messy. We're working on a pipeline that combines optical table detection with text-position clustering to get the best of both approaches.

Once shipped, this tool will be most useful for analysts and accountants who routinely receive financial PDFs they need to manipulate. For everyone else, Ask PDF can often answer table-style questions directly from the document without needing to extract the table.