Bank statement extraction software

What data extraction actually has to get right

Pulling text out of a statement is the easy part. What makes bank statement extraction hard is that the numbers have to still be correct at the other end, and there are only three ways they go wrong:

1. Direction. Most statements print withdrawals and deposits in two unsigned columns. Read them naively and half your transactions import with the sign inverted — income where spending should be. You find out at reconciliation.
2. Dates. 01/02/2024 is 2 January in one country and 1 February in another, and nothing in the file says which. Guess wrong and a year of transactions lands in the wrong months.
3. Dropped rows. Multi-page statements, wrapped descriptions and carried-forward totals all give a parser chances to lose a line silently. A missing transaction is invisible; a wrong one at least looks odd.

Every amount is checked against the running balance

This is the part worth paying attention to. When your statement carries a running balance, ConvertFin does not trust the column headings — it reads the signed amount from how the balance actually moved between rows, then re-checks every row against that balance and tells you on screen whether they all agree.

So you get an answer to the question that matters before the data reaches your books: did anything come out wrong? A green badge means every checkable row reconciles. An amber one names the rows to look at. No other check in this workflow is worth as much, because it is the only one that catches a plausible-looking wrong number.

Nothing is installed, and nothing is uploaded

This runs as JavaScript in the browser tab you already have open. For a practice that cannot install software on managed machines, that removes the IT conversation entirely. For anyone handling client statements, it removes the harder question: a statement carries an account number, a full transaction history and every counterparty, and the safest place for that file is the machine it is already on. Details, including how to verify it yourself, are on the security page.

Where it does not fit

Being straight about this saves you a trial:

• Scanned statements will not work. There is no OCR here. The PDF needs a text layer — if you can select the text with your cursor, it converts; if it is a photo or a scan, it does not. What to do instead is in scanned bank statement PDFs.
• There is no API. If you need extraction inside a pipeline rather than in front of a person, this is the wrong shape of tool.
• It reads statements, not invoices or receipts. The whole design is built around the date / description / amount / balance shape.

What comes out

One row per transaction — date, description, amount, balance — in whichever format the next system wants. Excel with real numeric cells for analysis, CSV for Xero, Wave and QuickBooks Online, QBO for QuickBooks Desktop, OFX for Xero, GnuCash and Quicken-family tools, and QIF for Quicken, Moneydance and MoneyMoney. If you already have a file in one of those formats and need a different one, the file format converters go straight between them.

Who this is for

Bookkeepers and accountants onboarding a client with a year of paper history. Lenders and mortgage brokers reading the last three months. Anyone reconciling statements a bank will hand over as a PDF and nothing else. It works with most banks because it reads the layout rather than a per-bank template, and it is free for the first few statements a day with no account.