What is OCR?

Optical character recognition reads characters out of an image or a scan. It is the first step for anything that arrives as a photograph of a document, which in practice means a lot of trade paperwork, delivery notes and supplier invoices. Modern recognition is good and it is not perfect, and quality falls off with poor scans, handwriting and unusual layouts. Any workflow built on it needs a defined behaviour for low confidence, which should be to flag rather than to guess.