Or maybe by the time they've fully made up their mind to return the fridge, it's often after they've first decided there is indeed a better fridge out there, and already chosen where to buy it in the day or two before they start a return. In which case the time to advertise to them would be the window after purchase and before initiating a return.
Maybe... etc.
I don't know for sure, having neither worked at Amazon nor tried to market fridges. But I have seen many examples of this sort of retargeting of people who've just bought something with ads of the same type of thing, where general intuition, even that of experienced marketers yet alone normal shoppers, made it seem like a dumb idea and yet the data actually did prove it was a great use of advertising.
So maybe your gut instinct is better than the ad conversion predictions Amazon's code is making, with all their data. But I'd be surprised.
I'm not sure as to their reasoning. But maybe this decision is in fact based on their data? They're a pretty data driven company
Most people, after all, don't really have the spare money to buy two fridges at once, so they're going to want their money back from the first one before they buy the second.
Once they've started the return process, then Amazon can start recommending new fridges to them.
Again: this is all data Amazon totally has access to. Whether they are failing to gather it, failing to recognize its value, or simply choosing not to use it when they know it's creating a lot of bad recommendations, is unclear.