1. Targeting is rarely as sophisticated as you think. Many campaigns are run wide and cheap (all users in a region, maybe some particular hours, etc). It's easier to optimize after this initial run based on actual performance rather than trying to figure out the exact people who would respond.
2. Some companies are also taking ad auction bidding in-house and running their own algorithms with their 1st party data, like upselling their own customers. This doesn't need any relevance matching since they know exactly who they're going after.
3. Targeting is not free. There is a continuum of price vs precision and high precision is rarely worth the costs, especially if the product itself cannot support those margins. Again this is why optimizing after a wide start is better than targeting upfront, and also why if you're trying to reach a very narrow audience it's easier to just send them email, or even direct mail.
4. People only notice the bad ads, not the good ones which they like or are influenced by. This is no different than complaining about bad CGI in movies when your average sitcom is 50% artificial but nobody notices because it just works.
5. Recommending movies as entertainment with cost in a subscription plan is nothing like finding relevant ads on the internet.
6. The adtech industry has 2 of the most valuable companies in the world and generates petabytes of data and billions in profit proving how well ads work. A random blog post by an outsider who has no idea how the industry works but claims it's all broken in direct contradiction to the data just comes across the same as a flat-earth conspiracy theorist.
>The adtech industry has 2 of the most valuable companies in the world and generates petabytes of data and billions in profit proving how well ads work
Targetting a demographic in terms of search or social media or television makes sense — the information is there simply by virtue using the platform. This is fine and well, and what the article agrees with.
Pulling it from demographic to individual by means of mass data collection is quite new, and unproven — facebook and google made their money before profiling was a major thing, and thus their wealth is not proof that its successful; they certainly believe theres mobey in it, as well as the advertisers and trackers, and are transitioning to it, but they aren’t proof that they wouldn’t be just as successful if they never transitioned.
Ads work. This has been shown by the last century of their existence in every medium they can fit in. Personalized ads, specifically those that are based on your non-current activities (eg buy on amazon and get served shopping ads when you later browse fb) are questionable.
Denying all the science and data which has proven the concept beyond any doubt to instead believe in the same old rehashed ideology? Sounds exactly the same.
A flat earth makes sense if you only spent 5 minutes thinking about it and have no background in or exposure to basic science.
Similar to this blog post. It makes sense if you read it, think about it for 5 minutes, and have no background in the advertising industry or the dynamics of effective/efficient advertising.
Not to mention that advertising and content recommendation engines for a paid service are each wildly different in their underlying dynamics and economics. But who cares about nuance anyway.
> A flat earth makes sense if you only spent 5 minutes thinking about it and have no background in or exposure to basic science.
Not that I particularly want to jump to the defense of people that believe in a flat earth, but they are not people that "only spent 5 minutes thinking about it". A lot of time has been spent trying to back up their beliefs and to form a coherent theory. They're wrong, of course, for many reasons, but it's incorrect and dismissive to call them intellectually lazy.
The core of flat earth and similar beliefs is often a legitimate reaction against the large established systems in society that, from their perspective, appear to be failing in obvious ways while lying to them regularly that everything is fine. The are recognizing a problem, and attempt to apply something resembling science to try and find answers.
> 6. The adtech industry has 2 of the most valuable companies in the world and generates petabytes of data and billions in profit proving how well ads work
But that's exactly the author's point. The ad industry works but not because it knows so much about you. Like many other commenters here, I can confirm that it always puzzled me how bad ad targeting is (or how nonsensical retargeting is, or how irrelevant recommendations are, etc.) despite what everyone is saying about how they all collect and mine data on our every move.
In my whole time on the Internet for more than 2 decades now, I can remember maybe only 2 or 3 ads that I mistook for genuine content and even clicked, that was on Facebook. Most of the other ads on social networks and elsewhere is trivial retargeting (remarketing) that fails 100% of time on me. This is not an exaggeration, remarketing never ever works for me, it's always stupid and irrelevant. Every marketing and growth professional will tell you that remarketing is the simplest thing that works, though probably not on people like me, but apparently it is pretty efficient.
Looking back now I realize that the author is right in that simpler hacks may be more effective than sophisticated ML-based algorithms. The truth may be in that, similarly to how Hollywood can sell stupider movies better despite that it can afford the best screenwriters in the world, we may be witnessing a similar "dumbing down" effect of the online ad industry. The fact that Hollywood has become a multibillion dollar realm today (and thriving!) whereas blockbusters become more and more predictable and simplistic over time, only tells us that the ad industry is probably heading in the same direction.
Oh and there's nothing new for me in "the dirty secret of the ML movement" the author mentions. That's quite an unpopular opinion today but time will tell.
Just because personal targeting is not used as often is not evidence it doesn't work.
Like I said, advertisers control their own campaigns and often start wide open to optimize, and they also don't want to pay for precise targeting. There's also a giant market for performance pricing is pay-per-click or other action. This means impressions are free so showing your ads to as many people as possible is the better approach.
If you want to see what personal targeting can do then you should look at lookalike modeling which finds similar people based on interests and behaviors and their propensity to carry out the same action. This technology has created many millionaires in the affiliate marketing world as campaigns would automatically keep finding similar people.
There are trillions of ad impressions and it's impossible for every single one to be perfectly tailored to you. That's just not how the industry works but business practices are completely different for technical capabilities. That's what this blog post and commenters do not understand, and your limited and faulty human memory is not somehow proof otherwise.
I'm sorry but you keep lecturing everyone here how big and sophisticated the ad industry is. Nobody denies it, no doubt there are billions in circulation and no doubt you can build something and sell it upstream to one of the giants (no matter how useful the giant will find it).
However I'm yet to see one good recommendation or ad shown to me. Like I said, very close to 100% of all advertisement shown to me, just as well as "friend" recommendations on social networks etc etc - all those things are so stupid and irrelevant that I refuse to believe there's something going on under the hood other than just plain stupid algorithms that probably work for some categories of consumers but not the others. The most relevant things happen only when I look for specific consumer products on Google and what I get is some ads in French which I don't even speak. Seems to me like tens of billions wasted. But of course capitalism is capitalism, they earn their money and they are free to spend it the way they like it for as long as it doesn't cross certain privacy rules in my country of residence.
Why do you say that "nobody denies it" when that is exactly what this blog post is doing?
You are claiming to remember all ads seen over decades. Even people with eidetic memory cannot do this, and in my experience people who claim to never see a perfect ad are the most susceptible to advertising. Influence is a lot more complicated than a simple banner ad that you think you've foiled by not clicking.
And I've explained several times why every ad impression is not perfect relevancy for you. What ads you see are a highly complex mix of the platform, running campaigns, targeting chosen by advertisers, optimizations in play, predictive analytics and propensity to action, pricing models, 3rd party data providers, inventory supply chains, creative formats, and many other factors. Trying to take trillions of ad impressions and derive the state of tech from it is both inaccurate and nonsensical.
Well your point 1 just confirms what the blog post said: you don't need to mine petabytes of data to get ads out to people. Point 5: so what? The post is about tracking, data collection and evaluation in general, not just ads. Butthurt ad industry guy? Point 6 doesn't really excuse all the smaller players responsible for the dozens of trackers on every news outlet. Google doesn't need to buy tracking data from anyone, so they don't have to solve the problem of correlating all that anonymized tracking data with questionable success.
Just because many ads do not use precise targeting does not mean that precise targeting does not work. That is the fundamental problem with this post.
Netflix recommendations are not the same as ad selection. You cannot generalize across such vastly different scenarios, datasets, and incentives, especially because relevance scoring is just a small part of what ad actually gets chosen.
I don't understand what you're saying about point 6 since this is not about Google vs smaller players, but there is definitely a monopoly problem with a single company having all the data.
Butthurt? I'm not 5 years old so no, however I do have more than a decade of experience in the industry, know the CEOs of all the major ad networks and publishers, personally presented to senators on increasing regulation, wrote about adblocking and built one to discover alternative payments, worked on finding and eliminating adfraud, helped build several successful marketing companies, and am willing to have open discussions with hundreds of comments right here on HN. Do you have some questions you would like to ask instead?
> Do you have some questions you would like to ask instead?
For example, regarding smaller players in the ad industry buying from a dozen tracking companies: does it really work? If yes it would either mean you have tremendously good algorithms to correlate anonymized data, or the data isn't really that anonymized to begin with. I mentioned google in the last paragraph because for them it's easy: they can track users better than anyone else and use it to show ads. It's all under the same hood. You dismiss the OP as a complete idiot, but doesn't it sound a least a bit likely that many smaller players just try to be google again here? Oh, google has so much data about the users to do ad targeting, we absolutely must do the same! There are so many places on the web where you have a pretty good idea about the demography of your visitors. Start from there. Most of the versatile places where you don't know who your visitors are are places like google, YouTube, Facebook, twitter, but they already know who their visitors are because they can do their own tracking, they don't need to buy any tracking data. So in the end I'm still wondering why there are two dozen trackers on CNN.com. who is buying all that data?
Late post but I didn't dismiss the OP as an idiot, I'm saying they are ignorant of how things work and extrapolating the state of technology based on visible end results is not accurate in anyway.
As for the rest, I've described this in the previous 2 posts. Precise targeting is possible. Every ad network has their own special focus, and yes many are useless or have been obsolete thanks to an evolving market. Not all of it is about a single visitor identity either. However just because targeting is available does not mean it's always used or always worth the price.
If you're selling toothbrushes, you don't need precise data. If you're selling million-dollar industrial equipment then it's worth paying the money to target the right people in their office. There are 1000s of factors that determine what you see and many are purely business and supply chain related with nothing to do with relevance so more often than not you'll see an ad that only has rough generic/contextual targeting and think the algorithm sucks when in reality nothing was applied in the first place.
2. Some companies are also taking ad auction bidding in-house and running their own algorithms with their 1st party data, like upselling their own customers. This doesn't need any relevance matching since they know exactly who they're going after.
3. Targeting is not free. There is a continuum of price vs precision and high precision is rarely worth the costs, especially if the product itself cannot support those margins. Again this is why optimizing after a wide start is better than targeting upfront, and also why if you're trying to reach a very narrow audience it's easier to just send them email, or even direct mail.
4. People only notice the bad ads, not the good ones which they like or are influenced by. This is no different than complaining about bad CGI in movies when your average sitcom is 50% artificial but nobody notices because it just works.
5. Recommending movies as entertainment with cost in a subscription plan is nothing like finding relevant ads on the internet.
6. The adtech industry has 2 of the most valuable companies in the world and generates petabytes of data and billions in profit proving how well ads work. A random blog post by an outsider who has no idea how the industry works but claims it's all broken in direct contradiction to the data just comes across the same as a flat-earth conspiracy theorist.