The OCR engine will recognize high quality laser generated output in electronic format (not scanned) with character accuracy as high as 98%.
Those received via paper that are laser generated and good quality will return accuracy rates up to 90%.
Poor quality (creases, folds, tears, marked-up, etc.) will result in much lower accuracy levels, decreasing to 20%.
Equally important as accuracy is discrimination - it is not enough to just be "accurate", you have to discriminate between items that have high confidence (and do not need an operator to look at), versus items that are wrong, or doubtful, where you do want an operator to view the results before committing next step in the process for workflow process - Transform Active Capture. Adding validations rules (e.g. – field is numbers only would eliminate discriminating if a character is a “1” or an “l”) will drive better overall accuracy.
The end result of all this will wind up in between – each customer has to assess the balance of all these factors based on their environment which will drive the overall accuracy rates.
Our experience with invoice processing is for customers to have an expectation being 70-80% on average, which can drive a lot of efficiencies.