Common upload mistakes

Uploading a personal bank statement while the business workspace is active will save all extracted rows as business transactions. There is no automatic reclassification — you must delete the affected rows and re-upload with the correct entity type.

Make checking the entity toggle the first step of every upload session. The current selection is always visible at the top of the Upload page.

Photographs of receipts taken with a phone camera often have distortion, shadows, glare, or low contrast. These reduce OCR extraction accuracy significantly. The parser can only work with what OCR extracts — it cannot recover text that the OCR engine cannot read.

Password-protected PDFs cannot be opened by the processing system. Remove the password protection before uploading. Most PDF viewers allow you to save a copy without password protection.

Extraction is not perfect. Some transactions will have wrong categories, amounts with incorrect decimal separators, or dates missing from the parsed output. Skipping Review and relying on the raw extracted data for reporting or invoicing will produce inaccurate results.

Always run a full review session immediately after upload while the source file is still fresh in memory.

Numwisely checks every new file against what you already imported. If a file repeats an earlier one, for example because it was renamed, downloaded again or read slightly differently, its repeated transactions are skipped and the upload shows "Already imported". A statement that overlaps an earlier statement of the same account, for example January–June and then January–August, adds only the new months.

Review still flags possible duplicates that get through: rows with the same date and the same amount. You can also select Duplicates at the top of Review at any time.

A receipt and the bank line for the same purchase are flagged too, even when the dates differ. Your receipt shows the day you paid, but the bank often books the payment one to five days later. A receipt and a statement line with exactly the same amount in that window are matched. When the merchant or the category is also the same, the match is marked likely, and the bank line is kept: the receipt image moves onto it, and so does the receipt's category if the bank line has none yet. Two lines on one bank statement are never matched across days, because repeat purchases like a weekly fill-up are real.

In each match, select a copy to keep it or remove it. Usually you keep the copy you tracked yourself, or the bank line when a receipt is involved. Copies with different merchants from the same file start out all kept, because they are usually separate purchases. Nothing changes until you select Remove or Save, and you can undo the change afterwards. Decide later closes the review without changing anything.

Identical transactions can be real, for example two coffees from the same café on the same day. Select Not duplicates on that match, then select Save. Every copy stays and the match is not flagged again. If another copy with the same date and amount arrives later, the match comes back so you can check the new copy.

To remove a whole duplicate batch at once, use the source file filter in Review and delete its rows.

A card bill paid from your bank account is not a duplicate of the card purchases. It is matched as a card payoff instead and left out of totals, see "Card payoffs and transfers".

AI mode has a cost and a privacy implication. For most structured bank statement PDFs, Standard mode produces equivalent or better results. Try Standard mode first, then switch to AI mode only for documents where Standard mode performs poorly.