CSV vs PDF bank statements

Both formats land in the same review flow. How a CSV table and a PDF statement are each read, and which one to export when your bank offers both.

The same import either way

SavingStat reads bank statements in two formats: a CSV export or a PDF statement. Whichever you upload, the import behaves the same. You pick the wallet the statement belongs to, every row comes back as a draft with a category and a budget suggested, and nothing joins your records until you review and publish. Import a bank statement walks through those steps, and both formats pass through the same duplicate check, described in Find duplicate transactions.

What differs is everything before the rows appear: how the file is read.

How a CSV is read

A CSV is a table. Each transaction is a row, and once SavingStat has worked out which column holds the date, which the amount and which the description, every row is read from the table directly. That mapping is inferred once for a statement format and reused, so the same bank's export does not have to be puzzled out twice.

Because the values are read rather than transcribed, a CSV is the more literal of the two formats: what your bank exported is what lands in the drafts. A CSV export is taken up to 5 MB.

How a PDF is read

A PDF statement is a print document, so there is no table to read. SavingStat extracts the text and AI lifts the transaction rows out of it.

A transcription can drop a row, and a parse that is one row short looks exactly like a complete one. So the parse is checked against the statement itself: most statements print their own totals, such as an opening and closing balance or a total in and out, and the parsed rows are reconciled against those figures. When they do not line up, the statement is read again to recover what was missed.

A PDF statement is taken up to 10 MB, and it needs selectable text. A scanned statement that is only a photograph of a page fails with exactly that reason rather than being guessed at.

Which format to export

When your bank offers both, export the CSV. It is read directly rather than transcribed, so there is less room for a misread, and the file is smaller.

When the PDF is all you have, use it. The totals check exists precisely because transcription is fallible, and it catches the miss that matters most: a row silently absent.

Either way the rows arrive as drafts, and Review AI drafts covers what to look at before publishing.