Rowstamp reads the text that is already inside a PDF and rebuilds the table from where each piece of text sits on the page. When that text is there, the values are copied exactly, not guessed. The hard part is deciding which value belongs to which column and row, and that is where layouts can trip it up.
Works well
- Statements downloaded from online banking as PDF (text-based), with a table of date, description, amounts and balance.
- Statements spanning many pages, with the header repeated on each page.
- Descriptions that wrap onto a second line, and days where only the first transaction shows the date.
- Separate debit and credit columns, or one signed amount column (−12.00, (12.00), 12.00-, 12.00 DR).
- US, UK and European number and date formats (1,234.56, 1.234,56 and 1 234,56), Indian grouping (1,23,456.78), and amounts marked CR/DR or C/D.
- Layouts we test against on every change: US checking and credit card statements, UK statements, German (Buchungstag / Soll / Haben), French (Débit / Crédit), Spanish (Fecha valor / Importe), Brazilian (Histórico, daily "Saldo do dia" lines) and Indian (Ref No./Cheque No.) statements. These test files are generated by us to mimic each layout, so treat this as a strong guide rather than a guarantee for your bank.
- Invoices with a line-item table and a header row; digital receipts with one item per line.
Does not work (yet)
- Scanned or photographed documents on the free plan. Pro reads them with OCR in your browser. OCR is good on clean scans (our test scan converts with every cell correct) but can misread digits on blurred, tilted or faded pages. Always use the balance check on OCR results.
- Statements where columns touch. If two columns have no visible gap between them, the values can merge.
- Tables drawn as images or text converted to vector outlines (some "print to PDF" settings do this).
- Very free-form layouts: invoices where items are cards instead of rows, or statements that print each transaction across three or more lines.
- Generic tables with multi-line cells, merged cells or headers spread over two rows: you get the values, but some rows may need tidying.
How to check a result
For statements with a balance column, every transaction is re-added against the printed running balance. "45/45 rows reconcile" means every amount was read correctly and in the right column. Rows that don't reconcile are highlighted in red in the preview. You always see the preview before downloading.
Dates are kept exactly as printed (for example 03/04 stays 03/04), because a PDF doesn't say whether that is March 4 or April 3.