How to convert scanned bank statements to Excel or CSV
By the Banking Converter editorial team
Published
A scanned statement has no text layer, so it must be read with optical character recognition before it can become a spreadsheet. Scan at 300 dpi, keep the page flat and straight, run it through a converter whose OCR reports confidence per word, then verify the extracted rows with the statement's running balance. OCR mistakes on digits are common enough that verification is not optional.
Scan settings that read well
| Setting | Recommendation | Why |
|---|---|---|
| Resolution | 300 dpi (200 minimum) | Below 200 dpi, small digits lose the strokes that distinguish 3, 6 and 8 |
| Colour | Greyscale or black and white | Colour adds nothing for text and increases file size |
| Orientation | Upright, page edges parallel to the scanner | Skew is corrected automatically, but less correction means fewer errors |
| Format | PDF, one file per statement | Keeps pages in order and lets the converter treat the document as one statement |
| Phone photos | Even light, camera parallel to the page, no shadows | Uneven lighting is the main cause of OCR failure on photos |
What OCR gets wrong on statements
- Digit confusions: 0 and O, 1 and l or I, 5 and S, 8 and B, 6 and b. In descriptions these are harmless; in amounts they change the value.
- Dropped decimal points and thousands separators: 1,234.56 read as 123456.
- Column drift: on a skewed page, a value from the Debit column can land in the Credit column.
- Faint or struck-through text: bank stamps, hole punches and fold lines interrupt characters.
- Wrapped descriptions read as separate rows.
None of these are hypothetical; they are the everyday failure modes of financial OCR, which is why a spreadsheet produced from a scan should never be trusted without an arithmetic check.
Converting with confidence tracking
- Upload the scan to the scanned bank statement converter. The inspection tells you the pages are image-based.
- Convert. Pages are deskewed and rotated, OCR runs with word-level confidences, and the table is detected from the recognised words.
- Review the flagged rows. Rows containing low-confidence words are marked
LOW_OCR_CONFIDENCE; rows whose balance does not follow from the previous one are markedBALANCE_BREAK. Click a row to see the scanned line beside the value. - Correct and download. Edited rows are marked as user-modified and the exports are regenerated.
Why the running balance is your best check
A statement with a running balance column contains its own error-detecting code. For each row, previous balance + credit − debit must equal this row's balance. If OCR misreads 84.17 as 84.11, the next balance will be off by 0.06 and the check fails on that row. If it misreads the balance instead, the following row fails. Either way the error is localised to one line. The converter runs this chain automatically and reports the confidence score partly on its result.
Statements without a running balance can only be checked in aggregate: opening balance plus all credits minus all debits must equal the closing balance. That still catches most single-digit errors, but not two errors that cancel out.
When OCR fails
If the converter reports OCR_FAILED or NO_TRANSACTION_TABLE, the scan is usually too faint, too skewed or too low-resolution. Rescan at 300 dpi, or, if the original is a fax or photocopy, ask the bank for a fresh PDF. Some banks can regenerate old statements as text PDFs, which convert far more reliably than any scan.
Frequently asked questions
Can I convert a photo taken with my phone?
Yes, if the whole page is in frame, the light is even and the camera is roughly parallel to the page. Save or export it as PDF. Expect more flagged rows than from a flatbed scan.
Does OCR send my statement to a cloud OCR service?
Not with this converter: recognition runs on our own servers with Tesseract. Other services differ; check their subprocessors list.
Other bank statement converters
About the author
The Banking Converter editorial team is the engineering group that builds and tests the conversion engine: the parsers for regional date and number formats, the reconciliation checks and the CSV, Excel, JSON, OFX and QBO exporters. Guides describe what the software does and what the formats require; they are not financial, legal or tax advice.