I think it's much simpler. The chance that any person is say, buying a fridge at any point is very low. If you're 99% sure that person has already made their purchase, that means there's a 1% chance you're wrong, and they're still looking! That's a much better signal than any other signal you're going to get.
According to this article in the Toronto Star, online return rates are around 20%, rising to 30% in the holiday period.
https://www.thestar.com/business/2017/01/15/online-shopping-...
Actually, PC World had demo machines out to play with last time I was in there, but very few in comparison to the tech department of the department stores.
Maybe it's hard to connect Google's idea of who a user is with the store's user accounts. Maybe the ad-buying algorithm is suuper simple, and getting a dev to "do it better" isn't worth the cost.
If it's just "this person searched for car rentals two days ago, let's show them car rentals today," that seems simple enough to make money over "no targeting," with obvious and acceptable failure cases like "already booked a car" and "phone got stolen, should show ads for pawn shops instead."
However they also sell returned items to others. There are multiple huge warehouses where I live filled with (mostly) returned products being auctioned off to the public by a middleman. The volume is astounding.
There may be sense to this madness.
I believe rate of returns is increasing as online sales takes over. The only way to actually see and touch many electrical devices is to buy and return them e.g. there is no store in my major UK city that has premium non-mac laptops on display. It means you might buy and return 3 devices to find the one you want to keep. If the return rate is 30%+ then a completed purchase really is a strong indicator that might purchase the same thing.