Why can a bank statement PDF conversion miss rows or descriptions?
A PDF conversion can split or misread details when descriptions wrap, amount columns are spaced apart, pages repeat headers, a transaction crosses a page break, or OCR misreads a scan.
Why does a PDF split or reorder statement text?
A PDF stores positioned text, not always a clean transaction table. Descriptions may wrap to another line while amounts sit in separate columns; page headers and footers can repeat; and a transaction can continue near a page boundary. When a page is an image, OCR must also recognize small text, decimal points, minus signs, and punctuation.
These are layout and recognition risks, not proof that an extraction is wrong. They are reasons to review the converted rows against the source.
What should I check first?
- Statement period and dates: compare the first and last dates. If day and month could be swapped, choose the printed date convention before exporting.
- Descriptions: look for words moved to the next line, shortened text, or a continuation on another page.
- Columns and signs: verify that each amount belongs to the debit or credit column and that negative amounts keep their direction.
- Opening and closing balances: compare the statement values with the workspace summary and check any discrepancy.
- Page boundaries: compare the last transaction on one page with the first on the next to catch a split, repeat, or missing detail.
- OCR-sensitive values: inspect decimal points, parentheses, minus signs, and similar-looking digits on scanned pages.
Can a matching balance prove that every row was found?
No. A matching arithmetic trail is useful, but two mistakes can offset each other, and a missing row may not be detected if the source balance information is incomplete. Compare the row list with the source pages and investigate warnings even when the balance check matches.
ExtractMint lets you edit extracted rows and inspect the source PDF in the review workspace. It does not independently validate a bank’s statement layout; no bank-specific format has been independently validated against real statements.
Is this a scan or a text-based PDF?
Try selecting a transaction description in a PDF viewer. If it can be selected, the page likely contains embedded text. If it is a page image, OCR is needed. ExtractMint reads embedded PDF text first and can use browser OCR when the text is sparse; scan quality still affects the result.
See the scanned statement guide for scan review steps. The synthetic sample is invented and cannot establish accuracy on real statements.