It's completely bizarre to me that Amazon, with all that data and ML cannot detect this and hammer down.

Then again their best recommendations to me are extremely simplistic "you bought a washing machine, maybe you want a washing machine?" attempts, so I guess not.

Or they just don't care that much because that costs money and this is just easier and cheaper.

I've seen this brought up a lot, but if you split all your customers into people who have ordered washing machines in the last month and people who haven't, the first group is almost certainly more likely to be ordering a washing machine in the near future. Keep in mind that Amazon also knows the exact return rate for every item.

Even if 99% of washing machine buyers won't be in the market for a new one, that's probably still a better ratio than you'd get from advertising to random people who haven't bought one lately.

>but if you split all your customers into people who have ordered washing machines in the last month and people who haven't, the first group is almost certainly more likely to be ordering a washing machine in the near future

If, after 30 years of putting our best minds on the problem, this is the best we can do, then the entire advertising industry deserves to be burnt to the ground for making zero progress while wasting untold riches

I prefer to see it as a reminder of the omnipresent power of Bayesian inference :)

> Someone running a r(ating)-aas business might have 1000 accounts and order 1 cd per customer per account.

These are used Audio CDs. You're lucky if Amazon has one 3rd party seller with a copy to buy. There sure as hell aren't thousands available!

If they have 1000+ accounts ordering 1 CD per customer then this won't affect them until they have more than 4 customers at once.