> But ad algos don't care about conversion, they care about charging the ad buyer more.

Bingo. The users clicking the ads aren't the ones funding the advertising networks - it's the people buying the ads. If you can convince them that targeted ads are more effective (and it would certainly be a safe assumption based on most peoples' understanding) then you can charge them more.

I'm not sure why you're being so heavily downvoted for this, but the presence of so many people "in the industry" in this thread might be some indication.

Yeah, I've been struck by this possibility while reading a lot of the rebuttals here. "Targeted ads work because otherwise why would they be so prevalent?" or "targeted ads work because x, y and z." There could be a kind of meta-advertising at work here, where adtech is often actually marketing towards the ad market itself. Would that mean it works? In a way--clearly lots of people are buying into it. And people wouldn't necessarily notice the scheme, because after all even if targeting isn't as effective as they claim it's still a form of advertisement and would generate some sales as dumb ads regardless of whether the targeting is really effective. We'd have to measure how much above that baseline the ads are driving actual sales somehow.

But I've yet to see some actual controlled evidence showing clear data in terms of sales that we can be reasonably sure were generated by targeted advertising. I think such evidence probably does exist, I'm not saying I don't believe it just yet. But all the arguments here seem to confuse science with scientism--what we have is a lot of reasoning and academic explanations of how people think it works, which is just not the same as evidence. It doesn't matter if something sounds like the way it might work in real life, we have to make observations to call something data. I haven't seen much data here.

From professional and personal experince, remarketing ads, what you are describing, are much more effective ($ spent / products sold) for the advertiser than traditional ads.

So why do ads follow you even after you bought the product? Well the ads are still less than 1% effective (products sold / by ads shown), so for the average company (not Amazon) it is just not worth the engineering effort for a < 1% savings on your remarketing budget. Secondly, this is more speculation, but for many products the chance of buying a product given you have already bought one (e.x. a spare or a gift) is still higher than chance for a random person. As such the ads are still effective.

So what is the deal with the NYTimes? They a huge popular publisher (especially amount rich people) already command high ad prices through traditional adverstising. The value of a remarking ad (to the advertiser) is largely independent of the site that it is shown on. So The New York Times would naturally have fewer of these ads because the adverstiser gets more ads/$ on other sites. A diffrent site like a local paper might have a diffrent story. However remarketing ads have never been a large percentage of ads shown so from a revenue stand point publishers care a lot less about them.

That's not a good definition of effective. The advertiser actually cares about ($ spent / additional products sold that would not have been otherwise), which is much harder to actually measure as it involves a real causal question as to whether the ads influenced a purchase. Much of remarketing spend is likely wasted on customers who were going to buy the product anyways, especially so for people making a repeat purchase.

Its a not very well kept secret of internet advertising that it's so much more "measurable" than other types of advertising, but it's not actually measuring anything useful most of the time. There is the traditional quote that "I am wasting half my advertising budget, but I can't tell which half" With internet remarketing ads the answer might very well be both ;)

You mean advertising a Widget to someone who has just bought one? Yeah, I can see how that would be counter-productive.

I think even Amazon suggests other cellphone models "because I bought 'Y phone' recently"