Processes and understands scanned image documents using OCR, layout detection, and handwriting recognition.
The degraded case: no digital text layer at all, only pixels. Photographs of documents, faxes, old archives, handwritten forms. Everything must be recovered visually, and quality is whatever the capture device produced.
Archives, field-captured documents, and any process where the source is genuinely paper. Handwriting is the distinguishing requirement.
Where a digital original exists and could be obtained instead. Processing a photograph of a document you could have received as a file is avoidable difficulty.
Confidence thresholds must be set with the capture quality in mind, and low-quality captures should be rejected back for recapture rather than processed at low confidence.
Confident misreading of handwriting, particularly digits. A 7 read as a 1 in a monetary amount passes every check.
A restaurant group capturing handwritten till count sheets by phone photo at close of trade, where the count feeds the next morning's cash reconciliation.