What OCR does with a scanned statement
A statement downloaded from online banking carries a text layer, so every letter and digit can be read exactly. A scanned page or a photo is only a picture of text. Optical character recognition, OCR, turns that picture back into text, and DocFig then reads the transactions from it the same way as from any other statement.
Pages are detected automatically. A PDF page without a text layer, or an uploaded photo in JPG, PNG or HEIC, goes through OCR on the DocFig server, in memory, before the rows are read. Phone photos are turned the right way up using the orientation the camera recorded.
Why the balance check matters more for scans
OCR is good but not perfect. A faint 8 can be read as a 3, a crease can hide a decimal point, and a shadow can swallow a whole line. On a text PDF those mistakes cannot happen; on a scan they occasionally do. That is why every converted row is tested against the statement's own figures:
- Running balance. Each printed balance must equal the previous one plus the transactions in between, so one misread digit breaks the chain exactly where it happened.
- Opening and closing balance. The opening balance plus every row must arrive at the closing balance.
- Source check. Every amount must appear in the text read from the page.
When the running balance shows that money in and money out were swapped on a row, it is corrected and labelled. When a figure does not add up and the cause is not certain, the row is highlighted with the reason instead of being quietly changed. On a scan, expect the occasional highlighted row and treat it as a prompt to look at the paper.
Getting the best result from a scan or photo
- Lay the page flat and photograph it straight on, filling the frame, in even light without a shadow across the figures.
- Scan at 300 dpi if your scanner asks. Black and white or greyscale is fine.
- Include the page with the opening balance and statement period; those figures power the checks.
- For several pages, a single scanned PDF works better than separate photos, so the balance carries across pages.
- If the bank offers a downloadable PDF, use it: it skips OCR entirely and converts exactly.
What to check before you rely on the file
Read the summary at the top of the result. "Everything adds up" means the balances printed on the statement agree with every converted row. If rows are highlighted, compare just those rows with the paper; the fix is usually a single digit, which you can correct in the downloaded workbook. Statements that print no running balance can only be checked against their totals, so give those a quick look too. Scans are read with English OCR, so statements in other scripts may not convert well.
The output
The Excel workbook and the CSV have the same columns as any other conversion: date, description, money out, money in, a signed amount, the balance and a Check column explaining any flagged row. The file is held in memory only and nothing is kept once the rows reach your browser. See how files are handled. Have a downloaded PDF instead of a scan? Use bank statement to Excel.