What this actually does
Extraction is imperfect in every tool in this category, including this one. What is worth comparing is what happens next — whether the arithmetic is checked, whether you can see the line a figure came from, and what the software you use will actually accept.
Reconciliation that names the row
Five arithmetic checks on every conversion, from strongest evidence to weakest, with the failing row named.
Limit: What reconciliation cannot tell you
Every row points back at the line it came from
Select a row, and the page it came from is shown with the printed line marked — on digital statements and on scans.
Limit: On a poor scan the band is an estimate
Reading statements that are scans
Scanned pages are transcribed by a vision model on our servers, then put through the same reconciliation as everything else.
Limit: A scan is always less certain than a text layer
Eight formats, because your software only takes one of them
One conversion, eight export formats, with the format-specific requirements handled rather than left to you.
Limit: A format cannot fix what your software will not accept
Categorising transactions, and why we will not guess for you
Bulk-label selected rows and carry the label into exports that support one. No automatic guessing.
Limit: No rules, no learning, no memory between statements
What you will not find here
There is no API, and no team seats. Batch conversion exists for digital statements but not for scanned ones, which stay one at a time on purpose — each costs money per page to read and is worth watching. If any of that is what you need now, the comparison pages say who has it today.