Finance & admin

Extract table from PDF to Excel

Get PDF table entities: headers, rows, columns, page number, numeric cells, dates, footnotes, and repeated table sections.

Before you start

What to expect

Files are retained only for the product retention window and deleted after 24 hours unless your account policy says otherwise.
Uploads are not used to train models; AI Deep Read reads the page only to produce your requested output.
AI Deep Read costs 1 credit per rendered page and still needs human review for financial records.
A PDF table is reconstructed into a rectangle; merged cells, footnotes, and visual grouping may not match the original layout exactly.

One workbook containing tables AI Deep Read or the text-layer path can identify, with separate sheets when tables should not be merged.
Headers normalized where repeated, numbers typed as numbers, and dates normalized when obvious.
Page/table context retained so borderless or merged-cell tables can be checked against the source PDF.
CSV or JSON can be produced after review if your downstream tool does not need XLSX.

Input: 12-page analyst PDF with five data tables, two borderless tables, a repeated two-line header, and footnotes below each table.
Output: Workbook with five sheets, 312 data rows, typed Year/Revenue/Margin columns, footnotes preserved as final rows, and one merged Region header repeated down the group.

Everything lands in one column: Scanned or table-like PDFs route through Deep Read when needed; the workflow asks for tables rather than raw text dumps.
Merged cells and multi-line headers: Merged labels may be repeated or flattened so every row stays rectangular. Review header rows before using formulas.
Borderless or financial-report tables: AI Deep Read can infer many visual tables without grid lines, but blank spacer columns and footnotes may need manual cleanup.
Password-protected PDFs: Unlock the PDF first. The converter will not remove owner passwords or document restrictions.

Command line

camelot

bash

camelot -p all -f xlsx -o tables.xlsx report.pdf

Try lattice for ruled tables and stream for whitespace tables; neither handles scans without OCR.

tabula-py

bash

python -c 'import tabula; tabula.convert_into("report.pdf", "tables.csv", pages="all", stream=True)'

Good for quick experiments, but multi-table PDFs often need manual area selection.

Can it find more than one table in a PDF?

Yes. The workflow can output separate table blocks or sheets, which is safer than forcing unrelated tables into one grid.

Do digital PDFs use AI credits?

Not always. With Deep Read set to auto, PDFs with usable text layers can use non-vision paths; scans consume 1 credit per rendered page.

Will formulas be created?

No. Extracted tables contain values, not inferred formulas. Add formulas after you verify the rows and columns.

How are table footnotes handled?

Footnotes may be kept as rows or notes near the table rather than discarded, because they can change the meaning of financial values.