How to convert a bank statement PDF to Excel (and check it is correct)

Converting a bank statement PDF into a spreadsheet looks like a solved problem until you do it at volume. Then you notice the thing that actually matters: a converter that drops three rows out of two hundred produces a file that looks perfect and is wrong.

This is the difference between a tool that extracts and a tool you can trust with your books.

Why bank statement PDFs are harder than they look

A bank statement is not a table. It is a visual arrangement that a human reads as a table. Four things routinely break naive extraction:

• Multi-line descriptions. A single transaction can occupy three lines, and a parser that treats each line as a row invents transactions that never happened.
• Continuation pages. Headers repeat, running balances restart, and a page break can land in the middle of a transaction.
• Decimal conventions. 1.234,56 in Italy and Germany, 1,234.56 in the UK and US. Guess wrong and every amount is off by a factor of a thousand.
• Scanned statements. A photocopied or faxed page has no text layer at all, so extraction has to fall back to OCR — and OCR confuses 8 with 3 more often than anyone admits.

The check that matters: reconciliation

Every vendor in this market claims around 99% accuracy. The claim is unverifiable and therefore worthless as a way to choose between them. There is one check you can run yourself, on your own document, in a few seconds:

Opening balance + total credits − total debits = closing balance

Both balances are printed on the statement. If the arithmetic on the extracted rows does not reproduce the printed closing balance, something was missed, duplicated or misread — and you know it without checking a single row by hand.

Step by step

1. Upload the PDF. Machine-readable statements — anything downloaded from online banking — need no OCR and extract in seconds.
2. Check the reconciliation verdict before you look at anything else. If it balances, the row set is complete.
3. If it does not balance, look at the flagged rows rather than all of them. A good extractor tells you which rows it was unsure about.
4. Export to Excel or CSV, with dates in ISO format and amounts as real numbers rather than text.
5. For a recurring feed, use the API so the conversion happens inside your own system instead of a browser tab.

What to do about scanned statements

If your PDF is a scan, extraction quality is bounded by the scan quality. Three things help disproportionately: scan at 300 DPI rather than 150, scan straight rather than at an angle, and keep the original rather than a photograph of a printout. A statement that reconciles from a scan is a good scan; one that does not is usually telling you about the input, not the parser.