Banks we have tested against

There is no supported-banks gate on this product. We do not keep a template per bank and match your PDF against it, so there is no list of banks that “work” and no list that does not. What we do keep is a set of real statements used to test the parser, and it seems more useful to publish that than a number we cannot substantiate.

What is actually in the test set

157 PDFs in total. Of those, 61 are statements in their own right — the rest are guides, training decks, regulatory model forms and educational material with a statement printed inside them, which are useful for a different test (checking we do not read a picture of a statement as though it were yours) but tell you nothing about bank coverage.

Of those 61, 41 files from 24 real institutions are listed below. The remaining 20 are synthetic statements, government files with the account holder redacted, regulator model forms and a credit union core-vendor template — real documents, but not issued by a named bank, so naming one here would be misleading.

Every file in the deposit and credit-card groups carries a human-checked opening and closing balance. That is what makes them a test rather than a demo: if a change to the parser stops one of them reconciling, the build fails.

The institutions

United States

Canada

India

Indonesia

Nigeria

What the three statement types mean

Deposit account — a current or savings statement, checked with the ordinary closing equation: opening balance, plus credits, minus debits, equals the closing balance printed on the page.

Credit card — checked with a different equation, because a card statement does not close the same way: previous balance, minus payments, plus purchases, plus fees, plus interest, equals the new balance. We keep the two groups apart and never average them into one score, because doing so would hide whichever of the two is weaker.

Image-only scan — a statement with no text layer at all, where every character has to be recognised before any table logic can run. These sit in their own group for the same reason: their results are not comparable to a digital PDF’s, and blending them would flatter the number.

The case that decided how this product works

One file in the set is a business statement from Access Credit Union, taken from a published sample. Read whole in one pass, the model returned zero transactions — a clean, empty, entirely plausible result. Our own local check disagreed: it found ink on the page with no text behind it and said so. On the second pass, re-rendered and re-read, a commercial loan account and five transactions came back, and the balance reconciled.

That file is why this product says “there is something here I did not read” out loud. A converter that had only run a balance check would have reported success on an empty table, because nothing was left to disagree with. It is also why the list above exists in this form rather than as a count: the useful question is never how many banks a converter claims, it is whether it can tell you when it has missed something in yours.

If your bank is not listed

Upload a statement on the free monthly allowance and read the balance check. That takes about a minute and tells you something no coverage list can: whether your statement, from your bank, in that layout, comes out right. If it does not reconcile, you will know before you have spent anything.