If I bought a vacuum I didn’t like from Amazon I’d return it thanks to their generous return policy. Targeting ads based on returns could make some sense I guess if they wanted to roll the dice on recouping losses.

That actually would be very smart targeting - especially given Amazon actually asks you for a reason of return. So they could target people that said this vacuum doesn't work or they need a different model, etc. by offering them similar one but from other manufacturer, etc. But I don't think this happens, and it's require developing much more sophisticated models than dumb "bought together" ones they seem to be using now.

But does that help Amazon?

Buying a vacuum, returning it, and buying a different one is less profit than just keeping the first vacuum, so I don't see why they'd want to encourage this.

I always just figured that was because it’s much easier to figure out that I was looking for a vacuum cleaner than it is to know that I’ve already purchased one and don’t need to be targeted any more. Especially if the “looking for a vacuum” signal gets spread further, to more systems/parties, than the “doesn’t need one anymore” signal does.

Maybe it's because the purpose is to make money selling ads, rather than to necessarily sell you the consume what you need?

ie these systems have multiple customers, you, and the ad or marketplace sellers, and the behaviour of serving up ads for things youve already bought is happening to milk the advertisers/marketplace sellers?

I would say that the more parsimonious explanation is that:

a) they've got a model that has learned that people who buy X tend to buy X again, and

b) they don't have a model that can properly distinguish between classes of X for which this is true and classes of X for which it isn't. They are aware of the problem and have made several attempts to fix it over the years but the results have never been sufficiently compelling to incorporate into the actual recommendation model in prod

Some years ago, I saw a presentation on Neo4J by a developer for a major British supermarket. (I think it was Sainsbury's.)

They wanted to print an appropriate coupon after someone paid for their shopping. "£1 off hamburgers next week" or similar, to get the person to return to the shop the following week.

They said a very important feature was to avoid giving a discount for a product someone has just bought or a closely related product. This would annoy customers who felt they'd missed out. Therefore, they were updating Neo4J with the just-made purchases and calculating the recommendation in the time it took to process the payment.

Maybe Amazon's chaotic mess of similar products makes this much more difficult than for a large supermarket, or maybe Amazon doesn't think the annoyance is a big deal.

» Maybe Amazon's chaotic mess of similar products makes this much more difficult than for a large supermarket, or maybe Amazon doesn't think the annoyance is a big deal.

I think amazon.com is no longer day one if it ever really was. I've seen at least one instance where they lowered price and refused to price match sale that hasn't even been delivered yet. I think they have data shows that not all people will go through the hassle to return and rebut and/or there's enough lock-in that amazon.com does not care about frustrating users.

Maybe the people who would be frustrated would go through the hassle to return and buy again. I mean if you have prime and you haven't opened the box, it is just two more days.

I think another factor that may come into play here is distinguishing between "products that serve the same purpose" and "related products" e.g. between "vacuum cleaners" and "vacuum cleaner accessories" (filters, bags, additional attachements...), the later likely being purchased in addition to the former. In this case it seems like there's probably a clear line but could certainly get blurry in some cases. If two products have some overlapping features, but also some unique ones, how different do they have to be for people to be likely to buy both?

I always loved when it would say "You bought No. 26 in a comic series, do you want No. 19 and 13?"

Yes, they're commonly bought together, but there's mostly, if not completely, a one-directional relationship-- someone buying 13 might want 14, 15, 16, but not the other way around.

This feels like it's just a matter of adding one more signed value to the "people who buy A buy B" record table... a weight indicating if the take is mostly A->B, B->A, or neutral.

Or, instead of the usual recommendation widget, just show a link to "All volumes in this series". Or even incorporate it into the website structure like YouTube playlists do.