Correlation [LWN.net]

Correlation

Posted Jan 27, 2022 18:37 UTC (Thu) by tialaramex (subscriber, #21167)
In reply to: Goodbye FLoC, hello Topics by kleptog
Parent article: Goodbye FLoC, hello Topics

In my circle of friends it is usual to manipulate Facebook's algorithm so that the constant advertisements are for stuff you don't mind in your field of view. For a large majority of those friends it turns out that "goth girls in lingerie" is both an advertising topic Facebook can be persuaded to show you and acceptable visual noise. Even for those friends who would plausibly identify themselves as "goth girls" I don't think lingerie is something they're likely to buy based on a Facebook recommendation, but it's an acceptable source of visual noise.

Right now though my Facebook adverts are 100% wall-to-wall adverts for an "innovation centre" which I think was built in my city in late 2019 and presumably is now realising that nobody wants to rent their office space and they never will again, oops. Facebook doesn't necessarily know I live here, but it does know I studied here, since that's why I have a Facebook account.

I didn't explicitly tell Facebook my gender, but interestingly Google correctly guessed my gender and *despite that* it spent months advertising women's shoes to me after I bought one pair because they are pretty (they're seriously very pretty, they're on a shelf in my living room). Not just women's shoes, but specifically exactly that same pair of shoes. Again, they're very pretty but nobody is going to buy two pairs. Advertising is algorithm driven and algorithms are dumb.

I think the evidence suggests that actually all the fuss over tracking based Internet advertising was a waste of time and money. $1000 spent on tracked Internet advertising won't actually be more effective than $1000 spent on, say, a billboard or a magazine advert, but tracking means you will get more advertising executives trying to *persuade* you that it worked because they have more tools to lie with. In particular, while the sales guy telling you that definitely somebody saw the billboard and then bought a brand new car is easy to dismiss, the guy saying the same thing about your Facebook advert has a chart that says so, how can it be wrong? Well, the same as the first guy, it's lying. That guy was on his way to buy the car when he passed the billboard, and he was reading Facebook right before buying a car too, it's not causality it's coincidence, and they've got more powerful tools to harvest coincidence online.

Why did I see all those shoe adverts? Because it's correlated. This guy saw your shoe advert AND he bought shoes. What do you mean which happened first?

to post comments

Correlation

Posted Jan 27, 2022 22:55 UTC (Thu) by rgmoore (✭ supporter ✭, #75) [Link] (2 responses)

The problem with ads continuing to chase you after you've already bought the product is notorious. My suspicion is that this is an especially bad problem when your decision to buy the product isn't affected by the ad at all; you just decided to buy it by browsing on their web site. The ad broker is tracking your surfing habits and knows you've been looking at a product. If you had decided to look at the product based on an ad, they'd know because they track clicks. They would probably know when you bought it, because they really want to be able to prove that their ads lead to sales.

But if you don't follow the ad to the product, the broker will only know that you've looked at the product; they won't know that you've bought it. Worse, the brokers keep exactly who they show which ad secret from the advertisers because knowing who to target is the core of their business. The last thing they want to do is to tell advertisers who is being shown which ads. So they keep showing you the ad in hopes you'll buy, all the while keeping it secret from the advertiser, who could tell them it's pointless because they've already made the sale.

Correlation

Posted Jan 28, 2022 9:29 UTC (Fri) by nilsmeyer (guest, #122604) [Link] (1 responses)

> They would probably know when you bought it, because they really want to be able to prove that their ads lead to sales.

Perhaps that piece of information is missing? Instead the algorithm assumed you didn't buy but were strongly interested in buying since the sale hasn't been reported back to the advertiser.

Correlation

Posted Jan 31, 2022 6:36 UTC (Mon) by rgmoore (✭ supporter ✭, #75) [Link]

Perhaps that piece of information is missing?

Yes, that's the idea. The point I was trying to convey is that the information isn't missing randomly. It's missing because of the way online ads work, which means it can't easily be fixed without deep changes to the system.