[jefftk]

> All of this money and effort to create something that has never scientifically been proven to work

There are parts of the online ads business that are very scientific. They run A/B tests to figure out what will make them the most money, and they're very good at it. Amazon, for example: say you look at a blender there and don't buy it. Amazon has cookied you, and can then target that cookie on other websites, showing you pictures of the blender trying to get you to come back and complete your purchase. They have very good metrics showing how much they should pay for these remarketing ads, and they make a lot of money from it.

> Each article on a newspaper already provides the context.

The article you're reading does give some information. There are a few kinds of articles that people are especially likely to read before spending money ("what kind of phone should I buy", "which credit card should I get", "what do I do if I've been diagnosed with mesothelioma") and ads on those pages are worth a lot. But if you look over the front page of the Washington Post ("Opioid death rates soared in areas where pain pills flowed", "At rally, crowd responds to Trump’s criticism of Rep. Omar with chants of ‘Send her back!’", "House votes to kill impeachment resolution", "This German city had few foreigners. Then refugees changed it in some surprising ways") or any other general interest publication you'll see that most articles are targeted at, and going to be read by, a very wide range of people.

(Disclosure: I work on ads at Google)

[user17843]

Aren't A/B tests super short-term and specific to single advertisers? In the end the money in the system is limited, and if one competitor achieves a high level of revenue via A/B testing, someone else probably loses that amount of money.

I wonder whether Ad-tech has seen a reduction in revenue after Apple introduced Intelligent Tracking Protection, as these users can not be tracked via cookies.

This source claims there was no significant dip for Criteo for example:

https://www.thedrum.com/news/2018/08/01/despite-apple-s-game...

[nerdponx]

The conspiracy theorist in me suspects that they only introduced cookie blocking after they were sure that other fingerprinting methods were reliable enough that they didn't need cookies anymore. They can pretend like they're "doing something" when blocking cookies in and of itself is another variable they can use to fingerprint you.

[JohnFen]

I read a number of martech and marketing sites (because it's good to know your enemy), and the discussions there strongly indicate that this isn't the case.

They've been spending quite a long time freaking out about the impending death of the cookie, and talking about ways to mitigate that.

[jefftk]

> Aren't A/B tests super short-term and specific to single advertisers?

They don't have to be short term. I currently have an A/B test I'm watching that I've been running for over a year, and other people I work with have some that have been running even longer. The reason to run a test long term, though, is either (a) you think the world might change such that your results won't remain valid or (b) you think a short-term measurement isn't a good estimate because of user learning. I'm not sure why you think a short-term measurement would be a problem in this case: a short experiment where an advertiser switched from personalized to pure contextual ads would lose a lot of money, and I don't see why you would expect that amount of money to decrease over time.

> In the end the money in the system is limited, and if one competitor achieves a high level of revenue via A/B testing, someone else probably loses that amount of money.

It's not zero sum, not at all! Let's say you sell board games. You advertise on board game review sites, but there's not that much traffic there because most people who buy board games spend most of their time on unrelated sites. If you figure out a new way to identify people likely to buy games and start advertising to just them you bid up the cost of advertising to those users, and displace, say, low-value belly fat ads. What happens? You start bringing in a lot more money, because you're selling more games. Some of that money goes to the sites you're advertising on, which now make more money because someone has figured out how to better advertise to their audience. And the belly fat advertiser makes less money, because the pool of ad space no one wants for anything has dropped slightly. But the amount of money they're losing is much less than the amount you and the site owners are gaining, because people are buying more board games.

> I wonder whether Ad-tech has seen a reduction in revenue after Apple introduced Intelligent Tracking Protection, as these users can not be tracked via cookies. This source claims there was no significant dip for Criteo for example.

I would guess Criteo isn't losing money because they've switched to tracking users via other means. They were caught [1] using HSTS supercookies, which led browser makers to remove HSTS as a tracking vector [2], and I suspect they've moved on to other methods. But (a) it's hard to tell externally since so much fingerprinting can be done passively and (b) I work for a Criteo competitor so I'm probably biased to think poorly of them.

[1] https://twitter.com/gothamresearch/status/942800208441827329

[2] https://webkit.org/blog/8146/protecting-against-hsts-abuse/

[tomjen3]

Showing me the products like that are so annoying, because google doesn't know I purchased them elsewhere and keeps showing them forever.

Please, after a reasonable time try to show me things people who have brought X typically buy after 2 weeks. As an example I would have paid for a curtain service to my new appartment, but google kept pushing (bad, paid) apartment search sites at me.

I now google products only in incongnito mode, since it get so annoyed by this behaviour.

Facebook decided initially that as I was single I must be interested in paid dating sites (even though, statistically I would be better of with a site with a larger audience), so please don't take this as an attack on Google, but as an attack on limited machine learning.