First, let me get one thing out of the way: not all batches need to be checked.

• If a manufacturer has great systems and processes, and if they have a history of complying with your quality standard, there is usually no reason to send quality inspectors to their factory.
• Similarly, if you can easily return the whole batch and it can be reworked quickly, you probably don’t want to spend resources on checking whether they did as promised.

Now, let’s assume you do want confirmation of product quality.

Should you perform a random AQL inspection or a 100% check?

When to only check a random sample?

By default, most buyers try to keep the volume of work low, by doing a random inspection.

And, in the vast majority of cases, for consumer goods, they follow the ISO 2859-1 (or ANSI/ASQ Z1.4) standard, set an inspection level, and set AQL limits.

The classic scenario goes as follows. A quality inspector (working for the buyer, or working for a QA agency appointed by the buyer):

1. Goes to the supplier’s factory
2. Counts the presented quantity, picks boxes & products at random
3. Checks the products and their packing for visual defects
4. Confirms specifications are met
5. Prepares a report that includes a pass/fail result

Who needs a random AQL inspection?

If you work with multiple suppliers and your quality standard accommodates 2-3% of defective goods, this is probably what you need to do.

When to do a 100% inspection?

From experience, I would point to three reasons some importers want a piece-by-piece check:

1. The cost of delivering defective products is very high – it is worth spending more time to sort out all the bad pieces. For example, empty shampoo bottles that will go on an automated line (which might be stopped for hours because of one deformed bottle).
2. There is a strong suspicion that a batch has quality issues and the supplier hasn’t done a good QC job — this is a classic after a first inspection failed and there was no reaction. Note that, if you know what the issues are, this is a focused check (sorting the goods by checking for 1 or 2 specific issues) and it can be done faster than looking for all types of issues on all pieces.
3. There is a risk that good products get replaced by bad ones after inspectors are gone — sorting good pieces into cartons that are immediately sealed makes this swapping game harder to play. 

All this has a cost. Over time, importers try to get their key suppliers to self-inspect more effectively. However, this is not realistic with all suppliers.

For example, many Japanese apparel brands still check all the pieces they buy in other Asian countries…

Should the 100% inspection be done in the supplier’s factory?

If the quantity is low and the product is simple, it can be done in the factory that made the goods.

If the work takes 5 or more man-days of work, though, the bill gets significantly higher and it might not be realistic.

What is the alternative? Having the batch of products delivered to another facility where the goods are placed on a line, checked one by one by trained operators, and (if accepted) are repacked immediately. It is more efficient and can be much cheaper, than sending people out.

One major advantage here is speed and simplicity. In many cases, what buyers want to avoid is this scenario:

• Too many defectives are found, the batch is rejected
• The supplier doesn’t communicate about their reaction plan or pretends they only found 1-2% of defectives
• An inspector comes in again, same findings, and another rejection
• The supplier keeps playing that game until the buyer can no longer wait and has to request shipment

The difficulty, here, is to convince the supplier to:

• Deliver the batch to another facility before receiving full payment
• Accept not to get paid for the goods that were sorted out as defectives in a ‘black box’ process they couldn’t witness

If these obstacles can be waived, it can be an excellent approach. Another benefit is, some labeling and branding elements can be affixed on the goods in that other facility, and the supplier will not know the distribution channel, the selling price, and so on.

Does this make sense? Have you seen this in practice, and how did it go?