[bavent]

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."

[reggieband]

> 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]

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.

[reggieband]

> Clearly I'm missing something in my logic

Your logic may be solid but you probably lack sufficient data. Plato was a pretty smart guy but he thought the four elements were Water, Fire, Air and Earth. His mistake wasn't intelligence or even flaws in his logic - it was missing data. If you begin your logic from one or more faulty assumptions then you will arrive at wrong conclusions regardless of how intelligent you are or how flawless your application of logic is.

I don't have the data either - but you should at least use logic to consider possible reasons why you are seeing the same products you have purchased previously. Perhaps this is a strategy that wins significantly more often than you assume and you just lack the data to illuminate why.

As the other commenter noted, you are but one data point in the literal hundreds of millions of data points available to Amazon. It would be humble to consider that they've tried your best first guess approach and it was suboptimal compared to the alternatives running now. Or maybe you choose to believe you are smarter than every single engineer that has worked on the problem there? And you believe this while lacking any data, having performed no experiments, etc.

[0xffff2]

>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 have the industry expertise to know how true it is, but I've seen it stated numerous times in threads like this that this is entirely intentional. Supposedly the data shows that a person who just bought a rowing machine is actually quite likely to buy another one (because they aren't satisfied with the first one, because they collect rowing machines, etc) compared to most other demographics.

[kd0amg]

> Clearly I'm missing something in my logic

Sampling methodology. Every ad impression in your sample was shown to the same person, but the ad industry is interested in billions of people.

[bavent]

I’m sorry, I don’t really follow. Could you explain a little bit more?

[cmorgan31]

You weren’t served a hyper personalized ad because those get expensive at scale so they segment your id into ad target groups. You fit some form of gym enthusiast who shows interest in rowing segment and some unfortunate rowing company keeps throwing their money at you in a useless bid.

[svachalek]

Is it one tiny increment better? Maybe? I'm with others who are not really impressed. For years the best Facebook could do were "Hot single [your gender preference] in [your city] are looking for [your age] [your gender]". Thanks, Mad Lib ads.

But for the sake of argument let's say there is some small increment. What is the cost we are willing to pay for that? Databases that know exactly how much time we spend on the toilet? Political disinformation campaigns? Insurrection attempts?

It's like making baby monitors a little bit more convenient, but in the process opening the door for pedo hackers to speak directly to your children's cribs. Tiny conveniences aren't worth sacrificing everything we have.

[reggieband]

The 1% chance you will click on a singles ad is worth more to Facebook than a 90% chance you will click on an ad for your favourite hobby. The singles company will pay higher for that 1% click chance than anyone else. Even if Facebook could show you a more relevant ad, they are not incentivized to do so.

It doesn't matter to Facebook/Amazon whether or not they show you ads you are interested in (or that might impress you with how deep they understand you as a person). It only matters if advertisers get better results than they would get by spending their ad money on TV, Newspapers, Magazines, Radio, etc.

[Terretta]

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]

Ah so it's the "Nigerian Prince" scam, basically?

[artificial]

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.