What does 99.6% accuracy mean in distribution centers? To the casual observer, it would mean that on average, 1 of 250 orders have errors. In market research, however, we have to look at the sample frame, or the source of the data compared with the total census of all instances. In this case, the sample frame is often customers (or other ship locations downstream of the DC) who complained or otherwise adjusted the order when it arrived.
Customers who did not complain could have been of three types: 1) customers who did not notice or otherwise care about the error; 2) customers who got the right amount of product, or 3) customers who got too much product and kept the overshipment for themselves. There might be lots of reasons for customers to keep over-shipments, including the cost of sending them back, the desire to make up for lost profits elsewhere, or even good old-fashioned five-finger discount (aka shrink). Nevertheless, the fact that these customers don't complain means that actual error rates are likely upwards of 1 in 250.
Hence the story my lead generation guy tells about a checking in on a customer who implemented Vocollect(R) Voice: his DC's downstream customers were very pleased with the improved accuracy, but they asked the DC manager, "What happened to all the extra stuff you used to send us?" The answer: the DC didn't mean to send it in the first place.
Improving accuracy means decreasing largesse for the downstream parts of the supply chain. In this case, that's a holiday bonus that isn't good for business.