When it comes to random inspections (usually conducted based on AQL tables), the inspector has to check a certain number of pieces. For example, if there are 5,000 pieces in the whole batch, he/she has to check 200 pieces (in normal severity, level II).

If the inspection was properly prepared, the inspector has certain checkpoints to follow. However, does it mean that these points have to be checked on all samples (in my earlier example, on 200 pieces)?

This is usually not realistic.

Let’s take an extreme example: an inspection of Android phone. Doing a full function check on 1 piece could take an entire man-day. So it is not realistic to check all functions (as per the instruction manual) on all samples, in this case.

Very often there are checkpoints that take more than a few seconds per piece. If the inspector checks all these points on the whole sample size, it can easily get the workload from 1 man-day up to 5 man-days. This is quite common.

To drive my point home, here are a couple of examples:

• Checking the size of the export carton — it usually makes sense to measure one carton, and to have a quick look at the others (“are they roughly the same size?”).
• Abusing a product for 5 minutes to see if it breaks or gets deformed easily — do this on 200 samples, and just this one test will take 1,000 minutes, or 17 hours.

So, in my mind, the logic should be as follows:

1. What we can check very fast (in the case of your product: visual defects, fitness test between lid and body…) is checked on the whole sample size.

2. What we can’t check very fast is divided in two categories:

2.a What is very important to the buyer (“critical to quality”), doesn’t take a very long time to check, and/or might vary substantially from one sample to the next due to the production process: can be checked on the whole sample size but it might increase the workload substantially.

2.b Other checkpoints: to be checked on a smaller sample size.

A few more remarks:

• In the 2.a case, any problem is classified as a defect, just the way a visual defect is counted. It makes sense since all inspection samples were checked for this point (and the statistics from ISO 2859-1 were respected).
• In the 2.b case, let’s say a point is checked on 5 pieces and one problem is found on 1 of the 5 samples. The inspector can simply consider this problem as critical, which means the checkpoint is failed. And then the buyer (or the quality manager) decides whether it is critical or not.
• This is a bit crude but that’s the way most inspection firms have been doing for a long time. The alternative would be to inspect all points on all samples, but usually that’s out of the client’s budget.