← All answers
How do you stop OCR from misreading invoice amounts?
By Vaibhav Vijay ·
Short answer: Read the amount twice: once from the figures and once from the amount-in-words line that Indian invoices carry, and accept the invoice only when both agree. Mismatches go to a person, and their correction is locked so the next scan can't overwrite it.
Which checks catch which errors?
Checks used by the Ekatra invoice tracker
| Check | What it catches |
|---|
| Figures against amount in words | A misread digit or a dropped zero |
| Invoice number pattern, EK-YY-YY-NN | Wrong files and duplicates |
| Lock on manual edits | A re-scan overwriting a human correction |
| Audit trail | Who changed which number, and when |
Why do amounts get misread?
Scanned PDFs, unusual fonts and the Indian lakh format, such as 1,84,000, all confuse OCR. A single wrong digit becomes a wrong reminder to a client, which costs more trust than a late one.
Why use the amount in words?
Most Indian GST invoices print the total in words as well as figures. The two are written differently, so OCR rarely gets both wrong in the same way. When they agree, the number is almost certainly right.
Related questions
What should the team do if a number looks wrong?
Report it instead of silently editing it. The edit is logged and protected, and the report shows where the extraction needs work.
Read next