I worked in this area of the company - it's actually quite a hard problem. For one thing, these are some of the oldest systems owned by Amazon - the team/servi

I worked in this area of the company - it's actually quite a hard problem. For one thing, these are some of the oldest systems owned by Amazon - the team/service/system diagram is a whole massive mess, so there's that initial part of just grasping all of the pieces that go into it, and everything else that comes with legacy architecture (assumptions, data flows, formats, etc.). Secondly, uhh trying to limit the amount of info I give - this link hints at the issue concretely enough - https://aws.amazon.com/solutions/case-studies/itemsandoffers/ https://aws.amazon.com/solutions/case-studies/itemsandoffers... - but there's some pretty complicated automated systems that handle relationships between ASINs, and basically it's those rules that get taken advantage of. And the rules are hard to fix because they do genuinely merge together a lot of duplicate ASINs/manage other relationships between ASINs, and then sometimes they end up merging unrelated stuff. And some of the rules are ML based, so you've got the whole recall/precision stuff. And the people talking about categories, there's like 4+ different systems for categorizing ASINs which all have separate use cases and taxonomies, so any answer that starts with "just do X" is probably not grasping the number of interdependent parts there are. Not saying I'm a fan of Amazon at all - I left for good reason - but these comments are just so incorrect about how Amazon works under the hood. No comment on if it's intended behavior, but I never saw anything that would indicate it was the case, and we had multiple teams actively working in this space.