Convert a Scanned Bank Statement
A statement that was scanned, faxed or photographed has no text in it, only a picture of text, which is why most converters return an empty file. This one reads those pages with OCR in your browser and gives you the transactions as Excel or CSV. The image is never uploaded.
OCR runs on your device · nothing uploaded
How a scan is read
Drop the PDF in as normal; there is no separate scanned mode to choose
Each page is checked for a text layer, and only the pages without one go to OCR
Those pages are rendered at twice their size and read character by character in your browser
The rows come out as transactions, with the running balance recomputed so a misread line shows up
What that means in practice
- Page by page, not all or nothing: a statement where only the middle pages were scanned uses OCR on those and the fast path on the rest.
- One OCR engine per document, started on the first page that needs it. A digital statement never starts it at all.
- Pages are rendered at twice their nominal size before reading, which is what makes small print and thin bank fonts legible.
- The balance check runs on the result, so a figure OCR read wrong usually breaks the running balance instead of passing silently.
- Nothing is uploaded. The engine, the training data and the page image all stay in the browser tab.
Why a scan is harder than a PDF from the bank
A statement downloaded from online banking contains the text itself, so reading it is a matter of putting the fragments back in the right order. A scan contains pixels. The numbers have to be recognised from their shapes, and a 3 that has been photographed at an angle under a desk lamp is genuinely ambiguous. Expect to check a scanned result in a way you would not check a digital one.
That is why the balance reconciliation matters more here than anywhere else. Opening balance, plus the credits, minus the debits, should land exactly on the closing balance, and when OCR misreads a digit it usually does not. A mismatch tells you which statement needs a second look, and roughly how large the error is.
Scanned statement questions
Do I have to tell it the file is scanned?
No. Every page is checked for a usable text layer and only the pages without one are sent to OCR, so a mixed file with some scanned pages and some digital ones is handled correctly without you sorting it out first.
Is the scan uploaded to be read?
No. The OCR engine is WebAssembly and it runs in your browser tab, the same as the rest of the conversion. The page image, the recognised text and the transactions all stay on your device.
Will a phone photograph work?
Often, and it depends mostly on the photograph. Flat, evenly lit and square to the page reads well. A photo taken at an angle, in shadow, or with the page curved will lose characters. If you have the choice, a flatbed scan or a scanner app that flattens the page is worth the extra minute.
How accurate is it?
Good enough to save the typing, not good enough to trust unread. Treat the output as a draft you check, and start with the balance reconciliation: if the opening balance plus the transactions lands on the closing balance, the amounts are almost certainly right. If it does not, the gap points at what was misread.
Why is a scanned statement slower to convert?
Because recognising characters from an image is far more work than reading text that is already there. Pages are read one at a time on purpose, since running several at once is how a phone runs out of memory mid-conversion. A digital statement of the same length is close to instant.
What if my statement is a photo rather than a PDF?
Put the images into a PDF first, which every phone scanner app and both macOS Preview and Windows Photos will do. The converter takes a PDF, and one PDF containing all the pages is also how you get a single spreadsheet out rather than one per page.