Bank statement parser

Bank statement parser

Every bank lays out its statement differently, and half of them arrive as scans. Templates break on the next bank. This parser reads each page as it is, turns every line into a transaction, then checks the running balance so you can see what to verify.

Goes in

Bank and credit-card statement PDFs, digital or scanned, up to 20 MB

Comes out

Transactions as CSV, Excel or JSON, from the web app or the API

10/07/2026  CHK 1043 RENT - OAK STREET          500.00        4,766.61
{
  "date": "2026-10-07",
  "description": "Rent - Oak Street",
  "credit": null,
  "debit": 500,
  "balance": 4766.61,
  "reference": "CHK 1043",
  "classification": "OAK STREET RENTALS"
}

From a statement line to a transaction

An illustrative line as it might appear on a statement, and the transaction the parser returns for it. Column hints from the page decide which side the amount goes on.

Statement line (illustrative)

10/07/2026  CHK 1043 RENT - OAK STREET          500.00        4,766.61

Parsed transaction

{
  "date": "2026-10-07",
  "description": "Rent - Oak Street",
  "credit": null,
  "debit": 500,
  "balance": 4766.61,
  "reference": "CHK 1043",
  "classification": "OAK STREET RENTALS"
}

The full statement comes back in the same shape as the JSON export, or as CSV and Excel.

Sample data from a made-up statement, not a real account.

Who this is for

  • Finance teams and bookkeepers who receive statements from many banks and cannot keep a template per layout.
  • Developers who need a statement parser behind their own product, through the API or MCP.

How it works

  1. Step 1

    Read the page

    Digital pages are read as positioned text, keeping columns aligned. Pages with no text layer go to OCR.

  2. Step 2

    Extract rows

    A language model turns the page text into transactions: date, description, credit or debit, balance, reference, payee.

  3. Step 3

    Check balances

    Each running balance is recomputed from the previous row. Mismatches are reported, not hidden.

  4. Step 4

    Export

    Preview, then download CSV, Excel or JSON, or fetch JSON and CSV from the API.

What the parser does, and does not, do

It is a statement parser, not a general document parser. Knowing the edges saves you a test upload.

Parses

Bank account statements (checking, savings, current) and credit-card statements, digital or scanned, multi-page when signed in.

Does not parse

Financial statements such as balance sheets, income or cash-flow statements, and it does not read photos directly. Other account statements (brokerage, loan, utility) are not what it is built or tested for.

No bank list

It is not template-based, so there is no list of supported banks to check. The quickest test is a free upload of your own statement, no account needed.

Accuracy and limits

  • PDF only, up to 20 MB per file. Photos and screenshots (JPG, PNG) are not accepted; save them as a PDF first. If the PDF is password-protected, save an unlocked copy.
  • Without an account you can convert one PDF of up to 3 pages a day. A free account converts up to 5 pages a day; Pro 400 at $15 / month (up to 400 pages per month) or Scale 1000 at $30 / month (up to 1000 pages per month).
  • On our ground-truth benchmark the current engine scores 93% transaction F1 and 99% date accuracy. That benchmark is three hand-checked statements (Canara, HSBC, ICICI), so treat it as a sign, not a guarantee for your bank. Check the preview before you download.
  • The running balance of every row is recomputed from the one before it. A row that does not add up is flagged in the JSON (balance_check), not silently fixed.

Questions

What is a bank statement parser?

A tool that reads a bank statement, usually a PDF, and returns its transactions as structured data: one row per transaction with date, description, amount and balance, ready for a spreadsheet, accounting tool or code.

Which banks does it support?

It is not template-based, so it does not keep a bank list; it reads each page's layout. Our ground-truth benchmark covers three hand-checked statements, so test yours with a free upload.

Can it parse financial statements like a balance sheet?

No. It parses bank and credit-card statements. Balance sheets, income statements and cash-flow statements are a different document and are not supported.

Does it handle scanned statements?

Yes. Pages without a text layer are sent to OCR first. Photos (JPG, PNG) are not accepted; save them as a PDF.

How do I know the parse is right?

Two checks: the preview on the page before you download, and balance_check in the JSON, which lists the rows whose running balance does not add up.

Is there a parser API?

Yes. Create a key at /developers and send PDFs to /api/v1/developer/uploads; results come back as JSON and CSV. See /docs/api.

Related