OCR accuracy is not the hard part

Extraction is a solved problem on clean documents. Everything expensive happens after the text comes out.