When I interviewed at Amazon a couple of years ago (for a team on the retail side of things, this was their bread-and-butter) I raised this very point with my interviewer, about my concerns over the data-quality issues I was having lately with item properties/attributes and miscategorization. My interviewer said the main problem was bad data coming from their Marketplace and other third-party vendors. - he said this is was why they recently (at the time) brought in the great minds and heavy-guns for machine-learning to fix all of these issues - and at the time (2016-2017) I noticed things were improving.
But lately - over the past year or so - I've seen things getting worse - I'm assuming because the problem of well-meaning but ultimately inadequate data from Marketplace sellers is now solved - but what remains is fighting bad data from intentionally incorrect metadata from malicious third-party sellers seeking to game the system - which we see today in spades with counterfeit sellers, for example.
Amazon's data-quality issue mirrors Google's perennial black-hat-SEO problem - I think it'll keep their teams occupied for decades at this rate.
(This is why I want to work on self-driving cars...)