To not replace a practice that works for Medicare numbers with one that is inappropriate, why not work to add a "dumb" AI which develops a rule set for each type of document. Where it met an anomaly, the original image OCR would be temporarily linked to a list to be resolved. With human resolution, other cases of the same kind could have the rule applied along with the first attempt and the image and any need for further conditions reviewed. While one can never be certain and flagging as an anomaly is better than a bad choice, this method allows Adobe OCR to be tuned to organizational needs and the diversity of its document flow. The AI might be found in the many forms of R Programming Language packages, prototyped there. And then the GUI built into Adobe Reader using code parallel to that in R.
To not replace a practice that works for Medicare numbers with one that is inappropriate, why not work to add a "dumb" AI which develops a rule set for each type of document. Where it met an anomaly, the original image OCR would be temporarily linked to a list to be resolved. With human resolution, other cases of the same kind could have the rule applied along with the first attempt and the image and any need for further conditions reviewed. While one can never be certain and flagging as an anomaly is better than a bad choice, this method allows Adobe OCR to be tuned to organizational needs and the diversity of its document flow. The AI might be found in the many forms of R Programming Language packages, prototyped there. And then the GUI built into Adobe Reader using code parallel to that in R.