How to extract invoice data from PDFs to JSON or Excel

In the Extractions dashboard, start a new batch: give the schema a name such as “Supplier invoices”, set its fields, then choose documents from your library or upload PDFs. A batch can hold up to 30 documents on Free and 100 on Pro. Documents run in the background, and one failure doesn't stop the others; retry failed documents from the batch page.

Review the results in one table. Click a cell to see its evidence, type a correction if needed, and approve or reject each document. Exports use your corrections and leave out rejected documents. Exports have one row per invoice; in Excel, line items go on a separate sheet.

Use the amount type for money so the currency code is captured separately, and the date type so every date comes out as YYYY-MM-DD. Add a short description with the label your suppliers actually use, such as “Amount payable” or “Toplam”. Mark fields you expect on every invoice as required, so a missing one stands out.

Invoices often show several similar numbers — subtotal, tax, total, amount already paid. Describe exactly which one you want; the quote next to each value shows which text the AI used.

Extraction copies what is printed. It doesn't check whether the invoice adds up, whether a tax number is valid, or whether you've already paid it. Dates like 03/04/2026 are ambiguous; when the document doesn't show the order, the date is read day-first and marked unverified for you to confirm.

Document text is sent to OpenAI's API, whose API data-usage terms by default exclude API data from model training.