How to convert a scanned bank statement to Excel or CSV (without OCR disasters)

Paper statements, faxed statements, statements photographed on a kitchen table by a client — sooner or later a scanned statement lands on your desk and the converter that handled every digital PDF suddenly returns garbage. Scans are a different problem, and pretending otherwise is how wrong numbers get into books.

First, check whether it's actually scanned

The ten-second test: open the PDF and try to select a line of text with your cursor. If text highlights, you have a digital PDF — the characters exist in the file, and conversion is a layout problem. If your cursor drags a selection box over an unhighlightable picture, you have a scanned (image) PDF — the "text" is pixels, and everything must be read optically before any parsing can begin. Mixed files exist too: a digital statement with a scanned first page stapled in by whoever assembled it.

Why generic OCR corrupts financial documents

Ordinary OCR was built for prose, where a misread character costs nothing — "tbe" still reads as "the". Statements are the worst possible input for that tolerance: dense tables, tiny fonts, and columns of digits where a single 8-read-as-3 silently changes an amount by five pounds, dollars, or thousands. The failure is invisible precisely because the output looks plausible — a column of numbers is a column of numbers. Skew, low-contrast thermal paper, and hole-punches through the balance column all make it worse.

The workflow that keeps the numbers honest

1. Scan well if you control the scanner. Flat, 300 DPI, straight — a phone photo at an angle costs more accuracy than any software recovers.
2. Convert with statement-aware extraction. Our converter treats scans as first-class input: AI extraction reads the document like a person would, and — the critical part — every date and amount is re-validated afterwards, with implausible rows rejected rather than guessed. Generic OCR-to-spreadsheet tools skip that validation entirely.
3. Reconcile against the statement's own summary. Every statement prints its opening balance, closing balance and usually debit/credit totals. Check the converted file against all three; if they match, the rows between them are almost certainly right. This is the single highest-value habit in statement conversion, and with scans it's non-negotiable.
4. Spot-check the ugly pages. If one page was creased or stamped, eyeball its rows specifically — damage is localised, and so are OCR errors.

When the scan is beyond saving

Some scans genuinely aren't recoverable to bookkeeping standard — fax-of-a-photocopy artifacts, missing column edges, thermal fade. The honest move is requesting a fresh copy: banks reissue archived statements (sometimes for a fee, which is still cheaper than wrong books), and even closed-account statements are usually retrievable within the bank's retention window. A converter that confidently returns numbers from an unreadable scan is a bigger danger than one that refuses.

After conversion: same as any statement

Once the scan is clean rows, everything downstream works normally — merchant identification, subscription and fee detection, and export to CSV, Excel, QIF or OFX for wherever the data is headed. For a stack of scanned statements (the estate-cleanup and audit-response scenario), batch upload converts the pile in one pass with cross-statement deduplication included.