Reviewing English and Urdu bank documents after OCR
Follow a fictional remittance form through OCR, field correction, and independent maker/checker review. See what to check before approving extracted data.
Compare extracted values with the original form
A scan can be clear enough to read while still containing a faint digit, a clipped field, or a label that OCR associates with the wrong value. Review the extracted fields alongside the source image before sending them downstream.
Source form
Transfer amount (رقم): PKR 15,000
Fictional sample for illustration.
OCR output
Transfer amount: PKR 75,000
The first digit was misread.
Reviewed value
Transfer amount: PKR 15,000
The checker confirms the correction against the scan.
Separate correction from approval
Extract the fields
OCR reads the scan and proposes structured values for the form. In this example, a faint digit causes the amount to be read as PKR 75,000.
Correct the amount
The maker compares the extracted value with the scan and corrects it to PKR 15,000. The original and corrected values are retained for review.
Check against the source
A separate checker verifies the corrected amount against the source before approving the record. The maker cannot approve their own work.
Decide what to do when a field is unclear
Ask the maker to flag unreadable fields for follow-up. If the source doesn’t support a value, obtain a clearer scan or send the record to an exception queue. Define which fields require a second check before approval.
When evaluating a document workflow, bring representative English and Urdu forms with sensitive information removed. Test how the workflow handles faint digits, missing fields, and corrections. Check that reviewers can trace approved values back to the original document.