Invoices arrive as PDFs. They should not have to be typed in again.
Upload a vendor invoice and it waits in Unprocessed PDFs until somebody turns it into a bill.
Vendor, invoice number, date, and amount are read from the document and pre-fill the entry form.
A scan with no text layer goes through OCR, so a faxed invoice is handled the same as a native PDF.
Correct where a field sits on one vendor's invoice and the next one from that vendor is read the same way.
The original travels with the bill, so an approver sees the document rather than somebody's transcription of it.
Extraction that cannot read a field leaves it blank for you to fill. It fails quietly into ordinary data entry rather than into an error.
Nothing in the AP process waits on extraction. Every field it fills can be corrected, and every invoice can be keyed by hand. That is deliberate: automation that stops the work when it fails is worse than no automation.