When I start building stuff, I'm going to make the serial number random and keep a database with serial<>manufacturing data. All to keep my Evil Competitors from gleaning any useful information!

You don't need a database, just a mapping function and a seed. I haven't seen anyone else do this, but I worked out a solution a while back.

It'd need to be a symmetric function that appears at least pseudo-random, any suggestions for suitable candidates? But I'd probably have a database with assorted data about the production anyway, so adding a random serial number wouldn't make much of a practical difference.

Any symmetric cipher (e.g. DES, AES, Blowfish, countless others) would work.

Ah, of course!

Sounds quite similar to the serial number format used on Apple products until recently. First 2 characters were the country / factory of manufacture, 3rd digit was the year, 4th and 5th were the week. The next three were the machines actual unique serial number, and the final three were a model identifier.

Seems to be pretty common to store this information in "plain text" in the serial number. I'm sure it has advantages over just giving every product a UUID and having a lookup in a database somewhere, perhaps in doing product recalls - you can specify a range of numbers that are affected?

This would be far more interesting if it were finding out information that wasn't already known.

A friend of mine ran a team at a financial institution that did stuff like this every day. For instance, looking for sequential transaction numbers on large e-commerce sites to estimate sales.

This particular bit of research sounds like it then could be used (in conjunction with some ongoing data-gathering) to predict sales / units manufactured in advance of Microsoft's quarterly reports.

That type of information is very valuable to the right parties, which is why my friend's team was quite highly paid.

This makes me want to look at serial numbers for all my stuff.