Scanned bank statement PDF to spreadsheet: what works and how to check it

Is your statement a scan or a text PDF?

Two PDFs can look the same on screen. One stores letters. The other stores a photo of each page. Run at least two of these checks, because one can mislead you.

File size and properties are only hints. A black and white scan can be small, and the Producer line can be empty.

Being able to select text does not settle it. A scanner or a PDF program can run OCR and hide the recognized text behind the picture. The file looks exactly like a plain scan, yet selecting and searching work. That text is a machine's reading of the picture, so paste a few numbers somewhere and compare them with the page.

Check every page. A PDF can mix text pages with scanned ones, for example when someone adds a scanned sheet to a text statement.

A PDF to text tool is one more test. The one on this site says "No text found" when a PDF has no text layer.

Why a spreadsheet cannot read a scan

A text PDF stores each character and its position. Copying, searching and a spreadsheet's PDF import all work from those stored characters. A scan stores one picture per page. The amount 1,250.00 is a pattern of dark dots, and the computer does not know it is a number.

So there is nothing to import. A Microsoft support answer from February 2026 says Excel's PDF import works best when the characters exist as selectable text, and that an image PDF returns an empty result. It adds that a report which was printed and then scanned or emailed often ends up as an image, and advises getting the original export instead.

To get numbers out of a picture, software has to run OCR, short for optical character recognition. It looks at the shapes and guesses which characters they are. The word to keep is guesses. It is why most of this guide is about checking.

Option 1: download the statement again

The bank's system created these numbers, so a fresh download from it is exact in a way a scan cannot be. Try this before anything else.

1. Sign in to online banking and open the statements or documents area.
2. Download the PDF the bank offers there. Do not print it and scan it, and do not photograph the screen.
3. On the transactions page, look for an export or download button. A CSV file is a plain table that a spreadsheet opens directly. If your accounting software imports another file type, choose that one.
4. Pick the exact dates you need, then run the select-text test on what you got.

Check what you get. A transaction export can differ from the PDF statement: it may leave out the opening and closing balance, or use other dates. Keep the PDF for the balances, because you need them for the balance test below.

If a CSV opens as one wide column, use the spreadsheet's text import and choose the separator that matches your regional settings. Once you hold a text PDF, a desktop converter that reads PDF text and checks the balances can turn it into rows.

Option 2: ask the bank or the account holder

Online banking may only go back so far, and paper statements were never online. Then ask for a file, not a copy.

• Phone, visit, or use the bank's secure messaging rather than ordinary email. Ask for the statements for your dates as PDFs from the bank's system, or as a transaction file.
• Say you do not want scans or photos.
• Ask whether there is a fee and how long it takes.
• For a client's statements, ask the client to download them from their own online banking and send the files unchanged.
• Ask to switch the account to electronic statements, so this does not happen again.

A message you can copy and adapt:

Please send the statements for the account ending 1234, January to December 2025, as the PDF files your system produces, or as a CSV transaction file. Please do not send scans or photos. Is there a fee, and when can I expect them?

Option 3: OCR, and what it gets wrong

If the scan is all you have, OCR is the way. PDF programs, scanning apps, OCR programs and online converters all do the same job: they look at the picture and write out text or a table. None is exact. Even on clear images, reading printed text is still not 100 percent accurate.

Small error rates add up on numbers. This is arithmetic, not a test of any product: if a tool reads 99 characters in 100 correctly, an amount such as 1,234.56 has eight characters, so about 8 amounts in every 100 would contain a mistake.

What goes wrong on bank statements:

• Digits. Shapes that look alike can swap, such as 3 and 8, 1 and 7, or 0 and O.
• Decimal marks. A dot or comma is a tiny mark. If it drops, 12.50 becomes 1250. If the two swap, 1,250.00 becomes 1.250,00, which many spreadsheets read as text. Countries differ on which mark is the decimal one, so set the tool and the spreadsheet to the statement's number format.
• Minus signs and brackets. Money out may be shown with a minus, brackets or its own column. A thin minus can vanish, and a payment turns into a deposit.
• Columns. OCR reads lines of text, and where one column ends and the next begins is a second guess. Amounts can slide into the wrong column, a description that wraps onto two lines can become two rows, and headers or "balance brought forward" lines can show up as transactions.

Better input gives better output. Scan at 300 dpi or more, flat and straight, with no dark borders or shadows from the page edge. A copy of a copy is harder to read than the original page. If the result is poor, rescan before you start correcting. If a tool wants pictures instead of a PDF, a PDF to PNG tool can turn each page into an image, but it cannot add detail the scan never had.

An online converter receives the whole statement: your name, your account number and every transaction. Read what the service says it keeps, and for how long, before you upload. Some tools mark the characters they are unsure about. Review those first, then run the checks below on everything.

Check the result with the balance test

The balance test is your safety net. A statement is a closed set of numbers: the opening balance plus money in minus money out must equal the closing balance. If your sheet disagrees, something was misread, and you find out before the figures reach your books.

In this made-up example one payment of 1,240.35 was read as 1,040.35. The sheet's money out is 200.00 lower, so the closing balance it calculates is 200.00 too high. For a credit card, take the previous balance, add purchases, fees and interest, subtract payments and credits, and compare with the new balance.

The size of the difference tells you where to look.

The rule about 9 is an old accountant's clue: when two digits swap places, the difference divides evenly by 9. For example, 72 and 27 differ by 45, which is 5 times 9. It is only a clue, because about one difference in nine passes that test by chance.

If the statement prints a balance after every row, add a check column. With the amounts in column C (money out as negative numbers) and the printed balance in column D, put this in E3 and copy it down:

=D3-D2-C3

Every row should show 0.00. The first row that does not is where the problem starts.

Spot checks, totals and what a match does not prove

A balance that matches makes hidden errors unlikely, not impossible. Two mistakes can cancel, and a wrong date or description never changes a total. So also do these:

1. Check five rows against the page image: the first, the last, the largest amount, one with a minus sign and one picked at random. Compare date, description and amount character by character. If one of the five is wrong, check every row.
2. Compare totals. If the statement prints total money in and total money out, or a total for each page, add the same columns in your sheet. Page totals narrow a mismatch to one page.
3. Count the rows on each page against the statement. Rows lost or doubled at page breaks are a typical conversion fault.
4. Check the dates. They should run in order and stay inside the statement period, with day and month the right way round. The opening balance of each statement should equal the closing balance of the one before.
5. Look for numbers stored as text. Amounts that sit at the left of the cell, or that the sum skips, usually hold a stray letter or the wrong decimal mark.

When to stop correcting

Fixing a few rows by hand is normal. If the test keeps failing and you keep finding new errors, the input is the problem: rescan with better settings, or go back to options 1 and 2.

Retyping a short statement is also an option, but typing errors are as likely as OCR errors, so run the same checks. If someone else will rely on the figures, such as an accountant or a lender, give them the bank's own file or ask what they accept. When the figures pass, matching them to your books is the next job, and the reconciliation template below is one way to do it.

Questions

Keep reading

More guides

About Ledgerpour

Ledgerpour is our Windows app that turns PDF bank and credit card statements that contain text into CSV, XLSX, QBO or CAMT.053 files, and checks that the balances add up before you import. It does not read scanned pages. Review the converted data before you use it in your books. It is not in the Microsoft Store yet.

Sources

• Microsoft Q&A: Can't import PDF using Power Query (answer of 2026-02-25)
• PDF software maker's blog (2007): Is that PDF searchable?
• Open-source OCR project: improving the quality of the output
• Wikipedia: Optical character recognition (Accuracy, Pre-processing)
• Wikipedia: Transcription error (transposition and the rule of nine)
• Wikipedia: Decimal separator

Checked 2026-11-03.