A big challenge is how set up QA so you're sampling more than zero images but less than every single one of them if you don't need to. A second challenge is to have some flexibility so you can be more rigorous for nasty sets of documents or new capture staff, or if you're starting to see an increase (or drop) in errors. 

If you're working with larger volumes and want a standardized, consistent, "generally accepted" methodology for how to structure your sampling for QA purposes, consider using something like the Military Standard Double Sampling methodology. Nothing in it is specific to the military, or any DoD standard, or government requirements for capture or QA. It just happens to be a straightforward methodology that seems to make sense. 

I'm going to summarize it. It's going to sound more complicated than it is because it's a statement of rules. But you design the rules into your process so no operators are working through a logic puzzle. Most of it is just automated.

The methodology involves random sampling of imaged documents, using defined sample sizes corresponding to various batch sizes. An initial random sampling is conducted, and if the number of defective units identified is less than or equal to the defined acceptance number, the quality review is complete. If the number of defective units is equal to or greater than the defined reject number, the entire population is rejected and a 100% inspection is instituted. Should the initial sampling yield a number of defectives between the acceptance and the rejection values, a second random sampling is conducted (excluding the sample items pulled for the original sample pool). In the second sampling, if the number of defective units identified is less than or equal to the defined acceptance number, the quality review is complete. If the number of defective units is equal to or greater than the defined reject number, the entire population is rejected and a 100% inspection is instituted. 

The methodology allows for Normal, Tightened, and Reduced levels of sampling, based on tracking of quality statistics. This allows for greater sampling of groups that fail quality checks and corresponding reduction when sampling shows improvement to the defined standards.

It's an Excel table. I can send it to anyone who wants it or you might find it online.

List archives at http://lists.ufl.edu/archives/recmgmt-l.html
Contact [log in to unmask] for assistance
To unsubscribe from this list, click the below link. If not already present, place UNSUBSCRIBE RECMGMT-L or UNSUB RECMGMT-L in the body of the message.
mailto:[log in to unmask]