We have invested heavily in building our own OCR Cognitive Service API that can reliably consume and process around 100k documents a day, extracting metadata against documents and populating various information against documents; To/From/Subject/Document Date/Author etc eliminating the manual aspect of document discovery.