Scanned bank statement OCR (Beta)
A scanned or photographed statement is just an image to a computer — no extractable text layer — so a normal conversion returns a blank table. Scanned OCR first uses a local tesseract engine to recognize the text in the image (with a digits-only character allowlist on amount columns and grayscale/threshold preprocessing to lift accuracy), then runs the same table extraction. We have to be honest: numeric and date OCR on financial documents is a well-known hard problem and measured accuracy is far below clean printed text — low-quality scans, handwriting and ornate fonts make it worse. That’s why this is Beta: after export, verify every single amount, and lean on the balance check to catch errors. Everything runs locally; nothing is uploaded.
A scanned or photographed statement is just an image to a computer — no extractable text layer — so a normal conversion returns a blank table.
Beta: financial-number OCR has limited accuracy and degrades with scan quality. Verify every amount row by row after export and rely on the balance check. If you can re-download a text-based PDF from online banking, use Bank statement to Excel instead — it’s far more accurate.
Scanned statement OCR (Beta)
- 1
Drag in a scanned / photographed statement PDF or image.
- 2
Pick the recognition language (English, or English + Chinese).
- 3
StatementSift runs local OCR + preprocessing (language pack downloads on first use), then splits rows and columns.
- 4
Verify every figure (really!), use the balance check to locate suspect rows, then export.
Scanned statement OCR (Beta)
Works even with no text layer: scans and phone photos that you’d otherwise retype get OCR’d into structured rows, sparing you most of the data entry.
Tuned for figures: a digits-only allowlist (0-9 . , - ( )) on amount columns plus image preprocessing push numeric recognition toward the usable zone.
Honest Beta + balance backstop: we don’t pretend a scan is as reliable as a text PDF — it’s clearly marked Beta, and the balance check helps you find the rows OCR got wrong.
Frequently asked questions
Honestly: less reliable than a text-based PDF. OCR does well on clean print, but numeric and date recognition on financial documents is a known hard problem with measurably lower accuracy; skewed, blurry, dark or handwritten scans are worse. That’s why it’s Beta — treat it as a tool that saves most of the typing, verify every amount by hand, and use the balance check to catch errors.
The OCR language pack downloads on first use (Chinese is ~20MB); your browser caches it, so later runs are faster. Overall speed also depends on page count and your device.
If you can re-download the official PDF statement from online banking (instead of scanning the paper copy), that’s a text-based PDF and Bank statement to Excel will be far faster and more accurate. Scanned OCR is the fallback when that isn’t possible.
Updated · StatementSift team
Sources, review and limits
Last verified
2026-06-17
Author
StatementCat editorial desk
Reviewer
Financial document workflow review
Primary sources
- User-provided bank statement files processed locally
- Local CSV, OFX and spreadsheet export rules
- Bank-format guides maintained by the site
Statement conversion is not accounting, tax, audit or financial advice. Reconcile exported rows against the original statement before relying on them.