Building a case for OCR for accounts payable is straightforward if you are honest about which minutes it removes. It removes transcription: reading a header off a document and typing it into a system. It does not remove matching exceptions, approval delays, or the judgement about a price that changed since the order, and a case that counts those as savings will not survive its first review. This page is how to build the case properly, using your own counts rather than a vendor's, and what to compare the quote against, because the honest number is usually smaller than the brochure and still often enough.
Separate the minutes into three piles
Transcription, which OCR removes. Routine handling, which capture reduces by feeding matching directly rather than through a folder. And exceptions, which it does not touch at all. Only the first two belong in the saving.
Use your counts, not published ones
The free approval worksheet on this site turns your invoice count, touches, minutes and exception rate into hours and money. On the worked example on this site, 400 invoices at three touches of six minutes with a 12% exception rate is 134.4 hours a month, $4,300.80 at a $32 loaded rate and $10.75 an invoice. Your figure will differ, and it is the only one worth putting in a case.
Compare the quote against the routine portion
If exceptions are 12% of invoices and take proportionally more time, the addressable part of the cost is smaller than the total. A case built on the total will overstate the return and be found out in the first quarter after implementation.
Questions people ask about ocr for accounts payable
What if the numbers do not justify it?
Then say so. Capture is often bought for consistency and audit trail rather than for hours, which is a legitimate case as long as it is the case being made.
Should we pilot it?
With your worst invoices, not your best. A pilot on the clean regular suppliers proves the easy half and tells you nothing about the exception path you will live with.
How long does implementation take?
Longer than the demo suggests, mostly because supplier layouts have to be learned and your coding rules encoded. Budget for a parallel month rather than a switchover.