Scanned statements

Convert a scanned bank statement to Excel

Scanned statements are images of paper. There is no text to select, so copy-and-paste returns nothing and a normal PDF parser sees a blank page. Converting one means recognising the characters first, then rebuilding the transaction table from their positions.

FlowParse detects pages without a usable text layer, renders them, runs OCR, and parses the result. Because OCR can misread a digit or a decimal point, scanned pages are marked and uncertain rows are flagged for review rather than exported as if they were certain.

  • Automatic detection of image-only pages, with OCR only where the text layer is missing.
  • Page-level OCR confidence so you know which pages need a closer look.
  • Flagged rows for low-confidence text, ambiguous dates and amount problems.
  • The same Excel and CSV exports as a digital statement, after review.

No card required. 3 processed pages free every month.

How it works

  1. 1

    Upload the scanned PDF

    Upload a PDF scan of the statement. Keep the scan straight and reasonably high resolution; 300 DPI is a good target. FlowParse rejects encrypted files, so remove password protection first.

  2. 2

    Let OCR and parsing run

    Pages without enough text are rendered and recognised. Progress is reported per stage, and the processor caps confidence for any statement that relies on OCR.

  3. 3

    Review flagged rows, then export

    Compare the flagged rows with the scan, correct anything OCR misread, confirm the reconciliation, and download the workbook.

Why scanned statements need a different pipeline

  • There is no embedded text, so extraction starts with optical character recognition.
  • Skew, shadows and low resolution change how characters are recognised.
  • Decimal points and digit shapes such as 3/8 or 1/7 are common OCR confusions.
  • Multi-column layouts can interleave if the OCR engine reads in the wrong order.
  • A confident-looking wrong digit is worse than a flagged row, which is why confidence is surfaced.

How FlowParse handles OCR

The processor checks the amount of usable text on each page. Digital pages are read directly. Pages that do not contain enough text are rendered at 300 DPI and passed to Tesseract, with word-level confidence retained.

Rows that come from OCR are never presented as a clean digital read. A mean page confidence below 65% marks the page low confidence, flags its rows and forces the statement's overall confidence to low. Any OCR page caps the statement at medium confidence.

Getting the best result from a scan

  • Scan at 300 DPI or higher, straight, with the whole page visible.
  • Prefer a greyscale or colour scan over a heavily compressed black-and-white copy.
  • Avoid photographing a screen; moiré patterns confuse OCR.
  • If the statement is available as a digital PDF, upload that instead. It is faster and more accurate.
  • Crop out irrelevant pages only if you are comfortable editing the PDF; otherwise upload the statement as issued.

What to check on a scanned conversion

  • Every page is represented, including pages where the statement repeats a header.
  • Opening and closing balances reconcile.
  • Rows flagged for low OCR confidence match the image.
  • Dates near the start and end of the period are correct, especially around month boundaries.
  • Decimal values look plausible for the account, not shifted by a factor of ten.

When OCR is not enough

A poor scan can produce a partial extraction. FlowParse marks the document as needing review or failed and lists the source lines it could not parse, so you can enter those rows manually instead of trusting a guess.

If a statement is handwritten, heavily creased or missing pages, OCR may not be reliable. In that case, request a digital copy from the bank or enter the missing transactions by hand.

Common scanned-statement situations

Clients send photos

A client photographs a statement and sends a PDF; OCR is the only way to reach the data.

Archive copies

Statements were printed and scanned years ago, and the digital original is gone.

Bank portals

The bank only provides an image-based PDF for a closed or legacy account.

Audit packs

A bundle mixes digital statements with scanned pages from another source.

Questions

Can I upload a photo of a bank statement?

Upload a PDF. If your phone produces an image, convert it to a PDF first, but a flat scan is more reliable than a photograph. FlowParse processes PDFs only.

How accurate is OCR on bank statements?

It depends on scan quality and layout. FlowParse reports confidence per document and per page, flags low-confidence rows, and marks OCR output so it is never presented as a certain digital read. Treat flagged rows as work to review, not as validated data.

Does OCR support every language?

English is the packaged OCR language. Statements in other languages may be recognised only where the text layer contains extractable text.

Will the export include scanned pages I did not review?

The export includes the reviewed ledger. If a page produced low-confidence rows, they remain in the ledger with review notes so you can fix them before using the numbers.

Is there a page limit?

The processor caps a PDF at 100 pages, and each plan has a monthly processed-page allowance. The Free plan includes 3 processed pages per month.

Convert your first statement free

The Free plan includes 3 processed pages per month. Upload a PDF, review the ledger, and export Excel or CSV when it looks right.