But even on this platform I've spent almost 10 years curating the perfect feed on, the one that should know absolutely everything about me, I still get served endless garbage ads for junk food and pickup trucks. I can't think of two products I care about less.
Amazon is a similar story, they seem convinced that I'm a woman despite having some pretty detailed purchase history going back to 1998 suggesting otherwise.
The targeting algorithm is often wrong, they just don't seem to know it be willing to admit it.
Advertising junk food and pickup trucks to someone that "can't think of two products I care about less" isn't an error. It's an attempt to make you care (if possible), and reinforce the brand so their name is familiar= if opinions change about junk food or pickup trucks in the future.
This sounds just like what people in the ad industry would tell worried clients. The actual reason is likely to be more mundane: the guy is, according to the tracking database, in an age and income group with high statistical correlation with interest in junk food and pickup trucks. Not all targeting is as detailed and personal as everyone thinks after reading scary articles.
Because the platform knows ads can be creepy, and will result in user disengagement.
Facebook often accurately targets people, and the result is terrified users.
So, the answer is other platforms have taken note, and are providing cover for users, to retain them.
This augments the return on ad spend, but only slightly, since ads actually aren’t very effective in practice. Ads are simply signage, nothing more. They don’t actually change behavior, but only reorganize behavior that usually would have happened anyway, so that the existing outcomes possibly land in the finite set of buckets differently.
Thus ad platforms only need to target so vigorously, and with that, it becomes obvious that targeting can be made less acurate, and buyers of ad space would never know.
So you can quickly see that Facebook’s game is absolute overkill, and other players likely underkill by a margin, with much improvement to their platform reputation.
But even if, doesn't this just bring you back to the original point of the blog post? Just advertise to everyone, no data mining and ML required. Best case, you do indeed change my mind about your brand, and the stereotypical junk food consumer sees the ad, suddenly feels hungry so hops into their pickup truck heading to the nearest McDonald's.
My interests, according to Facebook, include:
- "product (business"),
- "Providences and Territories of Canada",
- "Livestock", "experiment",
- "Lizard" and "Shark" (as a hobbies!),
They do get a few things correct, but none of these seem "inferred"--it's mostly organizations that I've explicitly liked and...substrings of their names.
I was reading a dark fantasy book series and all the similar items were romance novels…
It doesn't matter if their targeting algorithm is wrong. It just has to be 1) better than random and 2) not significantly worse than anyone else's.
In fact, even that isn't necessary. It just has to be perceived as those two things by those who are willing to give Google or Amazon money to try to improve their chances of selling you stuff.
If I want to target 18-25 year old males with make-up products then no algorithm is going to save me from myself.
I'm sure if your Twitter feed was full of automotive posts you'd see car ads. It's probably full of stuff no advertiser cares about, like your love of old British railway cars, so instead you get generic ads about spicy nacho cheese.
As a railway enthusiast myself, I would say that's actually an opportunity for very effective targeted marketing:
* There are loads of specialized reference and historic books and video on the subject.
* The crossover interest to model railways, a hobby involving many high-margin purchases, is very high.
* Many enthusiasts literally plan their vacations around preservation railways and museums. Everyone from the tourist-railpass sellers to full-fledged package tour operators should be targeting them.
If that interest is not well advertised to, it's either because they're doing a poor job identifying the interest and routing it to advertisers, or they've done a terrible job of fostering ad buyers in those verticals.
I've always taken pride, like I think most tech nerds of being "ad resistant."
However I've probably bought more stuff as a direct result of ad exposure on Instagram (3 things in three years) than in the entirety of the 25 years I've been on the internet.
Could you specifically describe the ads and the products involved ? I am very curious and would love to have a sense of what they succeeded with (for you).
I have been of the mind that targeted ads are terrible and that ad firms and ad-tech firms and (that entire ecosystem) are fleecing the ad buyers who will eventually wake up to that fact...
It's 2019 and if I search for "volkswagen emissions scandal" I will get banner ads for new volkswagen cars. For a week.
Or, I would, if ublock origin was disabled ...
I can only assume cause I've posted stuff about me cooking and have searched knife sharpening in the past. Not that hard to target really.
Ironically, the latter is a product that helps with decent retargeting, and one reason I purchased it, because it seemed to have worked on me.
I suspect the cuff one had a similar setup. I saw it once, abandoned a cart with it, and then maybe 30% of all ads I saw afterwards were the product.
Just as I'm realizing that it is something I'm quite interested in and want to click it for more information, the next page loads and I've lost the ad.
It's quite annoying.
In my case, I'm interested in climbing, and because brands actually post interesting content, I'll follow them. Do this over a few categories of interests and it may result in Instagram having a better understanding of my interests/things I'll care about.
The ads themselves are usually very high quality images or video, and despite taking up the entire screen like any other post, they don't feel intrusive.
Whatever they're doing is the right way to do it.
Instagram is all ads.
It's a platform where they get users to advertise to each other.
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.
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.
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.
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.
I recommend hbomberguy's recent investigation into recent flat earth belief: https://www.youtube.com/watch?v=2gFsOoKAHZg
I guess the flat earthers I’ve heard have been media personalities who had very few arguments to articulate.
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.
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.
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.
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.
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?
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?
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.
EDIT: this person worked as a software engineer at Google Fiber for 8 years, which confirms they have no experience with adtech.
This has actually nerfed search engines in my lifetime. It used to be a crawler returned as much as it could and you could search through the results; Now i get to only search through the big players who pay for their rankings.
>Never give positive feedback to an AI.
Funny; i wonder how Google profiling handles me occasionally using Google as a spell check.
I do that all the time!
If a programming placement firm makes $100 per lead (after various filter steps) on your targeting info, and a witty t-shirt makes $10, your cohort has to convert ten times as often on t-shirts to match the recruiting firm's ad bids.
For example, do you remember seeing ads along the lines of "if you or a loved one has been diagnosed with mesothelioma, call the law offices of such and so"? Mesothelioma affects less than 30 people per million. Why run an ad that so few people care about? Because the payoff is so high for reaching the people that do care, and the higher the payoff, the less people have to care for the ad campaign to be worthwhile.
On another note:
"But everyone sucks, except Pandora."
No, they suck, too. Thumb-up one song and it'll commandeer the station. Half the songs are live recordings (and I haven't checked if they finally added the option to exclude them, but given that it hadn't been added many years after the original feature suggestion by the time I switched to Spotify, I don't have high hopes); thumbing down said live recordings doesn't actually stop them from showing up (in fact, my "fuck this, I'm switching to Spotify" moment was when Pandora queued up three live recordings in a row, all of which I thumbed down, then on the fourth in a row wouldn't let me thumb it down because I had too many "skips" today).
Fuck Pandora. Spotify's radio feature is just as good (which ain't saying much, but it doesn't bombard me with live recordings, so that's a start), and of course Spotify supports use-cases besides procedurally-generated radio stations. Way more useful.
My guess is these targeters are maybe too sophisticated and rely on reams of data, social media logins, and etc. If you run a adblocker and clear cookies once in a while, you can "hide in plain sight".
Loved this part. About 99% accurate for "read this next" suggestions I've seen.
Personalization tech for ad tech is top of the line. Really pushing that part of ML forward. It drives the internet with billions worth of profit. Ad tech companies can know more about you than intelligence agencies, and sometimes they are one and the same.
Targeting is changing how people vote. It is influencing social mobility. It can turn startups into money printing machines. It is not something you can debunk in a single blogpost, just because it does not apply to you.
You are the vocal minority.
The key point in this blog post, I think, is the ineptitude of ML algorithms.
Computer vision can show greater-than-estimated-human-performance, while still failing hilariously unhuman once in a while. People remember the 1 in a 1000 wonky recommendation that made them do a double-take. Recommendation engines work best for the mean and stereotypical person. That way, you can use information of similar profitable people to effectively recommend.
Facebook, for instance, got mined for "suckers". If you are scummy, you want a list of gullible people who click the most stupidest, poorly designed, and shady ads. Ad tech knows where they are and delivers them on a silver platter. Going back 2 decades to serving ads without ML would kill a business. You don't think they thoroughly test a new recommendation engine and see relevant stats go up before they deploy it? You don't think they can serve you more relevant ads when they know you are a 17 year old male vs. a 42 year old woman? Both the data gathering and the algorithms have improved year over year. To say ad tech personalization is terrible, is akin to complaining we don't have AGI.
And yes, they test out their new ML algorithms to see how effective they are. But they don't need the machine learning to create their list of "gullible people." Rather, they don't even -need- that list. If you just serve stupid, shady ads that look as close to porn as that platform can get away with, you'd get your desired click-through rate. The actual personalization aspect is just a myth to validate our invasion of privacy.
Why would ad tech companies gather data and invade your privacy, and then not use it to sell more profitable ads? That makes no economic sense, but is a costly form of voyeurism. The "myth" you refer to sounds like a poorly thought out conspiracy theory.
Which is to say: personnalization doesn’t work.
This is just another meaningless "me too!".
This is just what advertising companies tell you, to sell you their product - targeted ads. It is not true to anywhere near the extent they would have you believe, for the reasons outlined in the post which are far more than just "it doesn't work on me". In fact, that's not really used as an argument at all.
He points out that tracking companies pay websites to allow them to collect this information, but (through examples) he argues that companies do not actually need the information that is being collected.
This begs some questions: Do these tracking companies remain solvent? Who is buying the information they collect? Are the buyers happy with the product/service? Is their usage of the data effective or just experimental?
I am willing to bet those tracking companies are operating based solely on investments, not licensing/sales of the data. That is, their future is uncertain.
We cannot put the genie back in the bottle. Whether or not the data works for the purposes it was collected, copies of the data will still exist. If the tracking companies fail, who gets the data then?
An ongoing effort will continue in trying to find a use for all this collected personal information by whomever shall come into possession of it.
How do you all think that will turn out? Should there be more regulation on how that data can be used?
BTW, another nice bit of work from this author which only got two points when it was posted on HN a few weeks ago:
lol, too true
Another thing here is that this information isn't going anywhere. All the information that you give up to companies can be exploited at their whim and to whatever end they choose. In this regard I think Cambridge Analytica was a really great thing. They were, all things considered, probably a very small player. But for people to realize that their personal information could be used for more than to try to sell them crap was an important lesson.
There are also things like the NSA who are now hoovering up immense amounts of information, facilitated by participating companies including Apple, Microsoft, and Google. Recent issues should show the problem here. The NSA is not immune to hiring people they should not (from their perspective) hire. Their secret tools get leaked. They themselves get hacked. And now there is this immense trove of potentially sensitive information that they're sitting on. That data is going to eventually end up in the wrong hands. It's also not out of the question that the NSA themselves eventually end up being the wrong hands. Without invoking Godwin's law here, it should suffice to say that bad people can get into positions of power and do very bad things. Pair this with profiles on everybody in the nation, and increasingly even the world, and it opens the door to some really catastrophic scenarios in the future.
So no, don't forget privacy.
 - https://www.sciencedaily.com/releases/2015/04/150408171201.h...
I'll watch a documentary about World War II, and then next thing you know it's recommending another 50 hours of Hitler and Nazi stuff. Or other war documentaries.
Medium is similar, too. I read a couple of articles about cryptocurrency, and then there's nothing but crypto articles. I ended up having to actively find a bunch of stuff to follow to get some variety.
I thought one of the key ideas in stats and ML was intelligent sampling? That would suggest you should sometimes throw in something the person hasn't expressed interest in, just to see if maybe you're on a local minimum. But I rarely see that.
I often wonder if it would be smarter for Netflix to just hire a guy who watches a lot of content, and he can just tell you what's similar to what.