Clioos › Guides › PDF to Excel

How to Convert a Scanned Bank Statement to Excel

By Clioos 团队

This guide covers the combination the other two guides split apart: a bank statement that is also a scan. The bank-statement guide assumes a digital download with a text layer; the scanned-PDF guide covers scans of any table. Here, both problems stack — the page must first be read by OCR, and then the result must still behave like a statement, with recognisable headers and a reconciliation that adds up. If either half fails, the spreadsheet cannot be trusted.

At a glance

First, find out which file you have: if you can select and copy text from the PDF, it is digital; if you cannot, it is a scan and needs OCR.
The no-code route is PDF Probe — first 10 pages free, no account, files under 10 MB and 20 pages or fewer, uploads deleted within 1 hour (live page, checked 2 November 2026). Scanned pages go through on-server OCR.
It only reads statement-style tables: Date, Description and Amount (or Withdrawals / Deposits), with a Balance column, and US-style dates such as 10/1/2026.
After conversion, the check that matters is the reconciliation: Beginning balance + all transactions must equal the statement's Ending balance. A difference other than $0.00 means a row was misread or missed.
Low-confidence rows are highlighted red in the preview. Verify every red row against the scan before you download — one wrong digit is worse than a missing row.

Steps

Check whether your statement is a scan: open the PDF and try to select a line of text. If nothing selects, the page is an image and needs OCR.
If it is a paper statement, scan it at 300 dpi, page straight, black text on white, even light. If you only have a photo, retake it straight-on, fill the frame, no shadows or glare.
Upload the PDF at PDF Probe (under 10 MB, 20 pages or fewer; first 10 pages free, no account).
Wait for OCR — longer than a digital PDF, usually a few seconds to about half a minute per page.
Check the preview: the reconciliation box (Beginning, Computed ending, Statement ending, Difference) and any red low-confidence rows. Compare red rows with the scan and check the row count matches the statement.
Download .xlsx. In Excel, re-check the running Balance column and the totals before you use the numbers anywhere else.

Worked example (local run, 2 November 2026 — not run through PDF Probe)

I built a synthetic one-page statement (invented bank and payees, 8 transactions, Beginning $1,000.00, Ending $2,408.14) in two versions and converted them with a local toolchain — Python (pypdf/pdftotext) on this machine. This example was not run through PDF Probe, so no number below is a PDF Probe result; the no-code OCR route is PDF Probe's on-server OCR, described on its live page.

Digital version (real text layer): pdftotext returned 569 characters of extractable text. A small parser (strip $ and commas, handle the $-1,200.00 sign form, keep US dates) recovered 8 out of 8 rows and 32 out of 32 data cells (8 rows x 4 columns: Date, Description, Amount, Balance) with 0 mismatches against the known truth file, 0 running-balance drifts, and a reconciliation of Beginning $1,000.00 + transactions = Computed ending $2,408.14 vs. Statement ending $2,408.14, Difference $0.00.

Date	Description	Amount	Balance
10/01/2026	Payroll - Example Corp	$2,500.00	$3,500.00
10/03/2026	Rent - Maple Apts	-$1,200.00	$2,300.00
10/05/2026	Groceries - FreshMart	-$84.20	$2,215.80
10/07/2026	Transfer from savings	$300.00	$2,515.80
10/10/2026	Utilities - City Power	-$112.50	$2,403.30
10/12/2026	Coffee - Corner Cafe	-$12.75	$2,390.55
10/15/2026	Refund - Online Store	$45.99	$2,436.54
10/18/2026	Pharmacy - HealthPlus	-$28.40	$2,408.14

Scanned version (same table rendered as an image, no text layer): pdftotext returned exactly 1 character (a form feed) — 0 extractable characters. That is the whole point of a scan: without OCR there is nothing to convert, however clean the picture looks. Scanned OCR was not run locally: tesseract is not installed on this machine (which tesseract returned nothing), so no scanned-OCR accuracy figure is claimed here. The scan file itself is also a clean computer rendering — no skew, no noise, no real scanner — so even a future OCR run on it would not prove how a real-world scan behaves.

What this example does and does not prove: it proves the digital half (extraction + reconciliation arithmetic) on a clean synthetic file, and it proves a scan carries no text to extract. It does not prove any OCR accuracy, any PDF Probe result, or any real bank's layout. Treat a real scan as guilty until its reconciliation reads $0.00.

Statement headers the converter recognises — and the ones it drops

OCR is only half the job. After the text is read, the converter still has to recognise the table as a statement. As documented in the sibling guides, the shapes that work are Date + Description + Amount (or Withdrawals / Deposits) + Balance, with US-style dates. The shapes that come back empty are: a table split into Debit and Credit columns instead of one Amount column; dates written as 2026-09-01; and generic non-statement tables. A scanned statement can therefore fail twice — once at OCR, once at header recognition — and the preview's row count is the fastest way to tell which half failed: 0 rows with a clean scan usually means the layout was not recognised, not that the OCR found nothing.

After OCR: the cleanup checklist

Row count first: count the rows in the preview against the statement. A missing row breaks the reconciliation even if every visible amount looks right.
Reconciliation second: Beginning + transactions must equal the printed Ending balance. If the Difference is not $0.00, a row was misread or missed — do not just accept the sheet.
Amounts must be numbers: after download, amounts should behave as numbers in Excel (SUM works, no text warnings). Strip stray $ signs, commas and spaces if you paste values by hand; watch the sign form — some renderers write negative amounts as $-1,200.00 rather than -$1,200.00.
Lookalike characters: OCR confuses similar shapes (& vs. semicolon is one documented case in the scanned-PDF guide). Read descriptions character by character on red rows; amounts and balances matter more than descriptions, but a wrong payee name still corrupts your records.
Dates stay US-style: the converter reads dates like 10/1/2026. If your statement prints dates another way, expect trouble and check the Date column cell by cell.
Do not burn free pages re-uploading: free pages are spent per upload even if the table comes back empty, so fix a bad result in Excel rather than uploading the same file again.

What it can't do

Handwriting: OCR reads printed text, not handwriting.
Skewed, faint or shadowed scans: a slight tilt is handled, but a photo at a sharp angle, with glare or shadows, usually fails. A flatbed scan at 300 dpi is the reliable route.
Heavily ruled black-grid tables: as documented in the scanned-PDF guide, a scan of a table with thick black grid lines returned 0 rows in testing; the same table with light grey rules converted.
Debit / Credit columns, ISO dates, generic tables: all return empty, scan or not — OCR cannot fix a layout the converter does not recognise.
No accuracy promise: nothing here guarantees your scan will convert, or that any single row is correct. The reconciliation and the red-row check are the verification; without them, do not use the numbers.
No indexing or ranking promise: converting a statement has no effect on search visibility, and this guide promises none.

Frequently asked questions

How do I know whether my statement PDF is a scan?

Open it and try to select a line of text. If text selects, it is digital and converts fastest. If nothing selects, the page is an image and needs OCR first.

Is the scanned-statement conversion free?

The first 10 pages on PDF Probe are free, no account — that includes scanned pages, which go through on-server OCR. Paid packs after that are one-time page packs; the live page lists prices from $0 to $79. Free pages are spent per upload even if the table comes back empty, so check the format before you upload.

My scan converted but the Difference is not $0.00. What now?

A row was misread or missed. Compare the preview row count with the statement, check the red low-confidence rows first, then compare the running Balance column row by row. Fix the wrong cells in Excel — do not re-upload the same file and spend another free page.

Why did my scanned statement return 0 rows?

Either the scan was unreadable (skew, shadows, faint print, heavy black grid) or the layout was not recognised (Debit/Credit columns, ISO dates, non-statement table). If the scan is clean, try your bank's digital PDF download instead — it converts far more reliably.

Is it safe to upload a real bank statement?

PDF Probe's live page states uploads and results auto-delete within 1 hour, files are never used for training, and scanned pages use on-server OCR — your file does not go to a cloud OCR service. If you would rather not upload a real statement, the format described in this guide is all the converter needs.