5 ms·

Does that actually work though? I've been buying things on Amazon for over a decade. For a while I got my groceries through them, I watch a lot of stuff on Prim

by bavent 6y ago

Does that actually work though? I've been buying things on Amazon for over a decade. For a while I got my groceries through them, I watch a lot of stuff on Prime Video. You would think they, of all people, know what I like. I can't think of more than a couple times where anything suggested to me was actually something I wanted. Even when I used FB, the ads were so off target from things I am interested in as to be laughable. Like, THIS is the best you can do with teams of engineers making > $200k/year throwing AI at everything? I'm not convinced that all of this tracking crap is just a way for them to market their ad business - "Look we gather all this data about people, your ads will definitely be seen by people who we know for a fact will be interested in them because of coding and algorithms and machine learning and blah blah blah."

artificial 6y ago
It comes down to the advertiser who is creating the audiences on the platform. "I can haz this many people?!" A terrible analogy is no programming language can save you from yourself. The current issue is the inaugural purchase mechanic, like buying a big ticket item or something that's a one off like toilet/toilet seat and now that's all you get moving forward.
Terretta 6y ago
Perhaps a small cohort of targeted-ad-susceptible users skew the targeting efficacy stats so far it looks like targeted ads work overall. Perhaps this cohort doesn’t overlap much with, say, New York Times readers, which might be why NYT and every other brand that tried first party non-retargeted ads saw an uptick in ROI. For most of us, perhaps these ads don’t work or are negative, while this cohort are more like Candy Crush in-app-purchase whales — for them they really really work, so they spend enough that most players think the game is “free to play” while griping about the endless in-app-upsells.
- bavent 6y ago
  Ah so it's the "Nigerian Prince" scam, basically?
reggieband 6y ago
> THIS is the best you can do with teams of engineers making > $200k/year throwing AI at everything? Maybe it is the best we can currently do and maybe it isn't objectively great. But the real question: is it better? I mean, better than the random shotgun approach that was TV, magazines, bus wraps, billboards, etc. Is it better than the spam flyers or "yellow pages" the post office delivered to every single household in some geographic area? Rather than compare to some idealized perfect, we should compare to the practical alternatives. Maybe this is legitimately the best we can currently do given the state of AI and machine learning. If that is the case, the right question for both advertisers and consumers is whether or not it beats the available alternatives. Because if it does, and advertisers seem to think it does, then that explains why Google and Facebook are worth what they are worth and how they can afford to pay what they pay.
- bavent 6y ago
  I don't buy that. For example - I bought a rowing machine on Amazon. It was the first piece of home gym equipment I'd ever purchased. For months after, other rowing machines were suggested to me. I already bought one, why would I buy another, especially if I hadn't returned the first one? Ads like this are very common for me - a purchase that I would consider to be a one-off thing (at least, for several years, whatever the standard lifetime of that product is) just leads to more ads for different models of that same thing. Occasionally, accessories that only go to a different model or brand of that thing. I don't see how this is any better than the random shotgun approach - these ads that are 100% irrelevant and not going to lead to a purchase are taking up space that ads that are possibly < 100% irrelevant (even if completely random) could be occupying. It seems like this would be a solvable problem for Amazon - aggregate the data of everyone who has purchased X model of rowing machine and see how many of them purchased a second rowing machine, which brand it was, and how long after purchasing #1 they bought #2. Don't show ads for rowing machines to people who have purchased X until time is >= avgTimeBetween1And2, with some fancy statistics in there somewhere. Clearly I'm missing something in my logic, because plenty of people a lot smarter than me work on this adtech stuff.
  - kd0amg 6y ago

by bavent 6y ago

artificial 6y ago

It comes down to the advertiser who is creating the audiences on the platform. "I can haz this many people?!" A terrible analogy is no programming language can save you from yourself. The current issue is the inaugural purchase mechanic, like buying a big ticket item or something that's a one off like toilet/toilet seat and now that's all you get moving forward.

Terretta 6y ago

Perhaps a small cohort of targeted-ad-susceptible users skew the targeting efficacy stats so far it looks like targeted ads work overall. Perhaps this cohort doesn’t overlap much with, say, New York Times readers, which might be why NYT and every other brand that tried first party non-retargeted ads saw an uptick in ROI. For most of us, perhaps these ads don’t work or are negative, while this cohort are more like Candy Crush in-app-purchase whales — for them they really really work, so they spend enough that most players think the game is “free to play” while griping about the endless in-app-upsells.

bavent 6y ago
Ah so it's the "Nigerian Prince" scam, basically?

reggieband 6y ago

> THIS is the best you can do with teams of engineers making > $200k/year throwing AI at everything? Maybe it is the best we can currently do and maybe it isn't objectively great. But the real question: is it better? I mean, better than the random shotgun approach that was TV, magazines, bus wraps, billboards, etc. Is it better than the spam flyers or "yellow pages" the post office delivered to every single household in some geographic area? Rather than compare to some idealized perfect, we should compare to the practical alternatives. Maybe this is legitimately the best we can currently do given the state of AI and machine learning. If that is the case, the right question for both advertisers and consumers is whether or not it beats the available alternatives. Because if it does, and advertisers seem to think it does, then that explains why Google and Facebook are worth what they are worth and how they can afford to pay what they pay.

bavent 6y ago
I don't buy that. For example - I bought a rowing machine on Amazon. It was the first piece of home gym equipment I'd ever purchased. For months after, other rowing machines were suggested to me. I already bought one, why would I buy another, especially if I hadn't returned the first one? Ads like this are very common for me - a purchase that I would consider to be a one-off thing (at least, for several years, whatever the standard lifetime of that product is) just leads to more ads for different models of that same thing. Occasionally, accessories that only go to a different model or brand of that thing. I don't see how this is any better than the random shotgun approach - these ads that are 100% irrelevant and not going to lead to a purchase are taking up space that ads that are possibly < 100% irrelevant (even if completely random) could be occupying. It seems like this would be a solvable problem for Amazon - aggregate the data of everyone who has purchased X model of rowing machine and see how many of them purchased a second rowing machine, which brand it was, and how long after purchasing #1 they bought #2. Don't show ads for rowing machines to people who have purchased X until time is >= avgTimeBetween1And2, with some fancy statistics in there somewhere. Clearly I'm missing something in my logic, because plenty of people a lot smarter than me work on this adtech stuff.
- kd0amg 6y ago