How to choose an AQL limit?

by Renaud Anjoran on 10 August 2012

A few months ago, I wrote an explanation about the “AQL“: what it is and how to use it. I also listed a few frequent questions, but it seems like I forgot one of them:

What AQL tolerance is suitable to my products?

Unfortunately, it is hard to base this on statistical reasoning. So I don’t have any definite answer.

The right AQL depends on two things, I think:

  • The market you are selling into,
  • The kind of risk the users run by using/consuming/getting close to your product.

1. The market

The most common AQL chosen by importers is 2.5% for major defects, 4.0 for minor defects, and 0.1 for critical defects. It is considered the “standard” tolerance for most consumer products sold in supermarkets in North America and in Europe.

Based on this standard, you can adjust an AQL that is a bit stricter (say, 1.0/2.5/0.10) if you sell your products in a high-end boutique channel. Or a bit looser (say, 4.0/6.5/0.1) for sale on a low-end market.

2. The user risk

For many car and plane parts, and for pharmaceuticals, the accepted defect rate is much lower than 1%. That’s because bad part might cause death. Specialists often use six sigma tools to reduce the opportunities for defects in every process along the manufacturing cycle.

I am not very familiar with these industries. I guess there are rules of thumb specific to each field.

To sum up:

There are no guidelines for deciding what AQL limits to choose. You have to decide what your tolerance. If the whole batch should not contain more than 1.0% of a certain kind of defect (over the long run), then the AQL should be 1.0% for this kind of defect.

Is it clear?

  • A KINGH

    great guy thanks

  • Eric

    Apparently the AQL is readily applicable to products that can be counted as “pieces” and “units”…How can it be applicable to products which are in (liquid) bulk and which are measured in terms of “metric tons”? Pls. help.

  • http://www.qualityinspection.org/ Renaud Anjoran

    Forget about this AQL in this case. Single stage sampling plans by attributes probably don’t make sense. You should probably go for a “accept on zero” plan: either what you test is good, or it’s all rejected.

Previous post:

Next post: