One person told us they nearly reported a real purchase as fraud because the charge showed up under a name they didn't recognize. That's exactly the problem Penny solves: she reads each line of your statement and turns cryptic entries into something readable, checking with you first whenever she's not sure, instead of guessing silently.
Account, card and IBAN numbers are masked before that text is sent to one of our two AI providers, Anthropic (Claude) and Google (Gemini). If the first does not return a usable answer, the other gets the same text, so either or both may receive it. If a statement's text cannot be read, the PDF file itself is sent instead, unmasked. Neither provider may use it to train its models. Anthropic deletes it within 30 days, unless its automated systems flag it as breaking its usage policy (it may then keep it for up to 2 years) or the law requires keeping it; Google keeps it for a limited period to detect abuse.
Anyone in the household can edit or delete a row she filed, and a correction teaches her how your household sorts things.