Accuracy Rate of OCR

The theme of our recent AP technology conference in Chicago was "Paper-Free AP," and several of the sessions addressed the use of data capture tools. That is, technology that reads the invoice once it's scanned into an imaging system so that the data does not have to be input manually. Several AP managers related positive experiences they are having with this technology.
Take a fresh look: We are aware that this technology, generally referred to as OCR (optical character recognition), has had somewhat of a troubled history. Some time ago, many people tried it and had bad experiences with it, especially in its accuracy rate.
But recently, we found out that users today are happy with the accuracy rate. Almost three-quarters of users we surveyed said that recognition accuracy met or exceeded their expectations. At our Chicago conference, users were reporting accuracy rates of over 90%.
Vendor claims: Vendors will quote accuracy rates approaching 97% or 98%, but those numbers are virtually meaningless. It's like anything you do in a laboratory under controlled conditions; the accuracy rate is totally different from when you're doing it in a real world environment. Published numbers have nothing to do with the appropriateness of the solution for your particular situation.
The only useful answer to this question is, "it depends on the kind of invoices you're scanning." The best test is to run images of your invoices through data capture applications from various vendors to determine which solution produces the best results. In some cases, it may be worth paying the vendors to install their systems for proof-of-concept testing at your location to obtain a real comparison of accuracy rates.

