We have to do manual validation for some of the things due to the OCR engine not being so accurate. For example, S is read as 5, 5 as S, 2 as Z, and Z as 2. In the case of currency, if one dot is missing, it gives a completely incorrect number. These discrepancies should be fixed because if the OCR itself doesn't work properly, then AI doesn't help with that. When AI is used for training models, the AI Center works precisely, but if the data itself is wrong, we can't do anything. If the extraction of data is proper, it definitely helps. However, 20% errors could be reduced if the OCR engine were better. In future updates, I would like to see better handling of issues such as when a dot is missing in currency, for example, with dollars, it should handle that automatically. It's very expensive. If the volume increases, we can't afford to use Document Understanding. We are thinking of moving to another solution because we are expecting 80,000 to 100,000 invoices.
