Run the five-second selection test
Open the statement and try to select a single transaction description. If the cursor selects individual words, it is probably a text PDF. If it selects the entire page as one object, it is a scan and needs optical character recognition.
Characters, words and table positions are available to the parser.
Pixels must be recognized as text before rows can be reconstructed.
What makes OCR succeed
- The complete page is visible, including column headers and page edges.
- Text is upright, sharp and large enough to read at normal zoom.
- There are no chat overlays, notifications, fingers or shadows over transactions.
- The screenshot has not been compressed repeatedly by messaging apps.
- Debit, credit and balance columns remain visually separated.
Cropped fragments remove the headers and balance context that help a converter understand what each number means.
Common OCR failure patterns
These are reasons to review the preview, not reasons to trust a polished spreadsheet automatically.
Improve a weak source before uploading
- Download the original statementUse the bank portal instead of photographing a screen whenever possible.
- Export at the highest available qualityAvoid social-media compression and low-resolution scans.
- Rotate and crop carefullyKeep all table headers and page boundaries.
- Test one pageConfirm that dates and amounts are recognized before processing the rest.
FolioParse flags rows and fields that need review before export.
Never treat OCR output as an official record
Keep the bank-issued file as the source of truth. The spreadsheet is a working copy for analysis, bookkeeping preparation or import. Use the conversion verification checklist before relying on totals.