Feature · Scanned statement OCR
Bank statement OCR for scanned PDF statements
A printed, signed and scanned statement has no text to copy. It is read by an extraction model, then checked against the statement’s own running balance, so a misread digit is flagged at its row before you export.
- Scanned and photocopied PDF statements
- The same five checks as digital statements
- Never used to train models
- Exports to all eight formats
The short answer
What bank statement OCR is, and what it has to do
Bank statement OCR turns the image of a scanned statement into transactions you can open in Excel or import into accounting software. Optical character recognition on its own only produces text; a usable conversion also has to find the transaction table, keep each amount in the right column, join wrapped descriptions and ignore the balance lines. VouchRow reads scanned pages with an extraction model that returns rows (date, description, amount, balance) and then reconciles them against the statement before you see them.
To tell whether your PDF is a scan, try to select a word with your cursor or search the document for a payee you know is there. If nothing highlights and the search finds nothing, the page is an image. Statements downloaded from online banking are almost always digital. Statements that were printed and scanned back in, faxed, photocopied or captured with a phone scanning app are images.
Digital statements are read in your browser. A scan cannot be, so its pages are sent to the processing layer for one request and then discarded, and the content is never used to train models. Scanned statements are converted one at a time.
Step by step
How to convert a scanned bank statement to Excel
- 1Prepare the PDFPut all pages of one statement into one PDF, in order. A password-protected PDF is fine: enter the password when asked; it unlocks the file in your browser and is never sent to us or stored. If you photographed the pages, save the photos into a single PDF first.
- 2Upload it to your accountUploading needs an account. When the file has no text layer you are told it will be sent for processing, and asked, before anything leaves your device.
- 3The pages are transcribedThe model is instructed to report what is printed and nothing else: merge wrapped description lines, skip brought-forward lines, headers and footers, transcribe the running balance, and settle day-first or month-first order once for the whole statement. Where a digit is unreadable it returns the row marked as low confidence rather than guessing a value.
- 4The result is reconciledThe same checks as a digital statement run: running balance row by row, page subtotals, opening plus movement against closing, printed totals and the transaction count. A misread amount breaks the running balance at the row where it happened.
- 5Review the flagged rowsClick a break or a low-confidence row and the scanned page opens with the line marked. Compare what was read with what was printed.
- 6ExportExcel with a Reconciliation sheet, CSV presets for Xero and QuickBooks Online, QBO for QuickBooks, QFX for Quicken, OFX, QIF, IIF for QuickBooks Desktop or Tally XML, depending on your plan.
Two paths, and which one your file takes
The browser opens the PDF and looks for a text layer. If it finds one, the whole conversion happens on your device and the file is never uploaded. If it does not, there is nothing local to work with, and the statement is sent to be transcribed.
You are told which path your file is taking before it happens, and asked. This is the one case where a statement leaves your machine, and it is not something to discover afterwards.
Why a model rather than traditional OCR
Classical OCR returns characters and their positions. That is a starting point, not an answer: turning a page of characters into transactions means working out which marks are a table, which column is the amount, which lines are a wrapped description, and which are a carried-forward balance pretending to be a transaction. Encoding those rules per bank does not scale past the first handful.
A vision model reads the page the way a person does, and the pages go up at their original resolution rather than through a canvas round trip that would lose detail before the model ever sees them.
The accuracy trade, stated plainly
A cheap fast model will misread digits on a poor scan. A dropped zero turns 220,000 into 22,000, and it will do this occasionally no matter which model is used.
That is survivable here for exactly one reason: every extraction goes through reconciliation before anyone sees it, and a misread amount breaks the running balance at precisely the row it occurred on. Take that safety net away and the model choice stops being defensible — which is why the two were built together rather than the OCR first.
What happens to a scanned file
It is sent for transcription and not retained for training. The request is routed only to providers whose terms do not permit retaining or training on what they receive, and a request that cannot be served on those terms fails rather than falling back to one that can.
What is stored afterwards is up to you: retention runs from keeping nothing at all through to a year, and expired data is deleted daily along with the stored PDF.
A scan is always less certain than a text layer
Why it works
What you get from scanned statement OCR here
Image-only PDFs read
Paper statements, faxes and photocopies that copy-paste and most PDF-to-Excel tools cannot touch.
Misreads caught by arithmetic
A 5 read as a 6 or a dropped zero breaks the running balance at that row, so OCR errors are found rather than imported.
Unreadable is not invented
A digit the model cannot make out is returned with low confidence and flagged for a look, not filled with a plausible guess.
Not kept for training
Scanned pages are sent for one request, only to providers that do not retain or train on what they receive, and then discarded.
Dates read consistently
Day-first or month-first order is decided once for the whole statement, so 03/04 does not mean March on one row and April on the next.
Same exports as digital
A scanned statement exports to every format your plan includes, with the same transaction IDs and the same reconciliation record.
Misreads
What OCR commonly gets wrong on a bank statement, and how it shows up
Every OCR engine and every model misreads characters on a poor scan. On a bank statement the misreads that matter are in the amounts and balances, and those are exactly what the running-balance check measures. A wrong amount does not hide; it produces a break whose size points at the kind of mistake.
The weak spot is text. The classic confusions between letters and digits, O and 0, S and 5, l and 1, B and 8, matter most in descriptions and reference numbers, and those are not arithmetic. A payee read as "C0STCO" balances perfectly. That is why the source-line highlight exists alongside the checks.
| Misread | Example | How it shows up |
|---|---|---|
| Similar digits | 5 read as 6, 3 as 8, 1 as 7 | Running-balance break; the difference is the gap between the two figures |
| Dropped or doubled digit | 220,000.00 read as 22,000.00 | Break of 198,000.00, a round multiple of the amount |
| Decimal point lost or moved | 1,234.56 read as 123,456 | Break at the row; difference is a factor of 100 out |
| Thousands and decimal separators confused | 1.234,56 read as 1.23456 | Break, or an amount far too small |
| Minus sign, brackets or CR/DR suffix missed | (45.00) read as 45.00 | Break of exactly twice the amount |
| Faint balance figure | Balance misread, amount correct | Two adjacent breaks, equal and opposite |
| Letters for digits in text | C0STCO, REF 5S01 | Not caught by arithmetic; check the source line |
Where it runs
Why scanned pages are read on a server, and digital ones are not
A digital statement carries its text and the position of every character. Reading the table from that is geometry, which a browser does well, so a digital statement is converted on your device. A scan carries only pixels, and reading a page of them accurately needs a model far larger than a browser can run. That is the only reason a scanned file leaves your machine.
The pages go up at their original resolution, are read in a single request, and are discarded. The same path is used, rarely, for a digital statement whose layout the browser parser cannot make sense of. How long the converted result is kept afterwards is your retention setting.
Scanned statements are converted one at a time, each with its own review. Digital statements can be uploaded in batches.
Scan quality
How to scan a bank statement for the best result
The model is not the limit on a poor scan; the image is. A few minutes spent on the scan saves more review time than anything else.
- Scan at 300 dpi. Lower resolutions merge the strokes that tell a 3 from an 8.
- Use greyscale or colour rather than pure black and white. A black-and-white threshold thins faint digits until they disappear, and the decimal point is usually the first to go.
- Lay the page flat and straight. Skew makes columns drift, and a curl at the binding blurs the right-hand side, which is where the amounts and balances are.
- Capture the whole page, edges included. A cropped balance column removes the strongest check there is.
- Scan the original if you have it. Every generation of photocopy loses detail, and a fax loses more.
- Keep pen marks off the transaction table. A tick beside a row is fine; a note written across the amounts is not.
- With a phone, use a scanning app that flattens and squares the page, avoid shadows and glare, and save all pages of the statement as one PDF in order.
Into your software
Getting a scanned statement into QuickBooks, Xero or Quicken
Accounting software imports bank transactions from files with structured rows: QBO or CSV for QuickBooks Online, QBO or IIF for QuickBooks Desktop, CSV or OFX for Xero, QFX or QIF for Quicken. A scan is none of those. Converting it produces one, and once it is converted a scanned statement is no different from a digital one: the same formats, the same transaction IDs, the same reconciliation record.
If your software’s own PDF import accepts your statements, use it for those. For paper statements, older history, closed accounts and anything your software will not read, convert first and import the file.
Mixed files
Statements that are partly scanned
Some files are mostly digital with one or two scanned pages: a page that was printed, signed and scanned back in, or a statement assembled from two sources. Those pages carry more risk than the rest. Check their flagged and low-confidence rows first, and look at the rows either side of where the scanned pages meet the digital ones, since page joins are where rows go missing.
Side by side
Digital vs scanned bank statements
| Digital PDF (has a text layer) | Scanned PDF (image only) | |
|---|---|---|
| How to recognise it | Text can be selected and searched | Nothing selects; search finds nothing |
| Where it is read | In your browser | Processing layer, one request, then discarded |
| Batch upload | Yes | One statement at a time |
| Typical errors | Layout issues: wrapped lines, page joins | Misread digits, signs and decimal points |
| Reconciliation checks | All five, where the statement allows | All five, where the statement allows |
| Source-line highlight | Exact | Located from the page image, or labelled as an estimate |
| Export formats | All formats on your plan | All formats on your plan |
Who uses it
When scanned statements turn up
Catch-up from paper records
A client’s statements arrive as a box of paper or a folder of scans. Convert each one, review the flagged rows, and import a clean history.
Closed and older accounts
When online banking no longer offers the period you need, the printed statement is the only record left.
Statements supplied by applicants
Lenders and brokers routinely receive scanned or photographed statements and need the figures in a spreadsheet they can check.
Disclosure bundles
Family-law and forensic work often runs on photocopied statements, where each figure has to be traceable to its page.
FAQ
Reading statements that are scans: questions
What is bank statement OCR?
Optical character recognition applied to a scanned or photographed bank statement, turning the image of the page into transaction rows (date, description, amount and balance) that a spreadsheet or accounting package can use.
How do I convert a scanned bank statement to Excel?
Combine the statement’s pages into one PDF, upload it, and let the scanned pages be read and reconciled. Review any rows flagged against the running balance or marked low confidence, then export Excel. The file has the transactions on one sheet and the reconciliation result on another.
Can you scan bank statements into QuickBooks?
Not as images. QuickBooks imports bank transactions from files such as QBO, CSV or, in Desktop, IIF. Convert the scanned statement to one of those first, then import it like any bank file.
How accurate is OCR on scanned bank statements?
It depends mostly on the scan. Clear, straight, 300 dpi pages read well; faint photocopies and faxes do not. No tool reads every scan perfectly, which is why every scanned conversion is checked against the running balance and totals, and a misread amount is flagged at its row.
Can I upload a photo of a bank statement?
The uploader takes PDF files. Save your photos into a single PDF, pages in order, ideally with a scanning app that flattens and squares each page.
Can I convert several scanned statements at once?
No. Scanned statements are converted one at a time. Digital statements can be uploaded in batches.
Is my scanned statement used to train AI models?
No. Scanned pages are sent for a single request, only to providers that do not retain or train on what they receive, and then discarded. A request that cannot be served on those terms fails rather than going elsewhere.
Does it work on scanned credit card statements?
Yes. Most card statements have no running balance column, so the check falls back to the statement balance, printed totals and transaction count. A misread charge is still caught, but it cannot be pinned to a single row.
Are password-protected scanned PDFs supported?
Yes. Upload the locked file and enter its password when asked. The file is unlocked in your browser, and the password is never sent to us or stored. The unlocked scan is then read like any other.
Keep going
Related guides and converters
Related features
Convert a scanned statement to
Who works with scans
Related
Other features
Convert a scanned statement and check it before you export
Create an account and upload a scanned PDF. Every row is reconciled against the statement’s own balances. Plans start at $15 a month for 300 pages, and only pages with transactions count.