My preliminary research on the Adtech marketplace

March 29, 2021

Note: You may notice that this post is dated older than when this website was created. That is because, this is a cross post. I had originally posted this on the Effective Altruism forum here. I am cross-posting it here in the spirit of collecting everything I have written in one place. You will notice there is a section that has been struck through - these were edits I made after reactions by readers in the forum. Clearly, not my best work! 🥲

What I would love to hear from you

In no particular order:

Link me to resources on how to potentially create an economic model of the Adtech marketplace. I would be grateful if you can direct me to any Economic concepts I need to look into to make a stronger argument.
Link me to relevant datasets.
Critique my understanding of how ads work. I am not a digital marketer myself and would like to correct any misunderstandings I have.
Critique my writing. This is my first “real” post in the EA forum and I haven’t written a lot in the past. So this is valuable to me.

Take everything that follows with a huge load of salt. Here goes nothing…

Context

As you may be already aware, online social platforms have caused problems like Information Disorder (spread of Mis/Dis/Malinformation) and data privacy violations among others. You may have watched the popular Netflix documentary ‘The Social Delimma’ about these problems or listened to the 80000 hours podcast episode where Rob Wiblin interviewed Tristan Harris. There seems to be a general consensus that these problems are occurring since these platforms are designed according to the needs of the underlying business model of Advertisement Technology (Adtech). This means that regulating the Adtech marketplace could help us address the other problems also. So in the last few weeks, I have been reading up on how the Adtech Marketplace works. Here are my learnings so far.

Why I think EAs should think about Adtech

NOTE: This section was added after tamgent’s comment on relevance to EA.

Given that online social platforms touch the lives of millions of people with the potential for more to join in the future, I assume it is fair to say the problems on these platforms like Information Disorder have a huge scale. But I find these problems quite intractable at a systemic level. It looks like a lot of work being done on these problem areas is to understand the problems case-by-case but not necessarily to solve them systematically. This report by First Draft or the work of Global Disinformation Index are good examples - they are building frameworks for managing the problem at a case-by-case level but not actually doing anything concretely to solve it systemically in a significant way. This is not to say that they are not doing valuable work. This could indeed be the best strategy for now. I sympathise with their approach since the problem is quite nettlesome. I feel conflicted to even suggest creating any policy directly aimed at this matter since letting the State regulate the Information Ecosystem directly has the potential to deter Freedom of Speech even in a Democratic Republic. So these problems seem highly intractable when approached directly.

But I believe an indirect approach might help. I think “The problem” of online social platforms is their engagement at scale. After all, Mis/Dis/Malinformation have been part of the Information Ecosystem long before online social networks were created, just not with the current level of engagement. There is evidence to suggest that this high engagement is a direct result of the design of these platforms to attract & sustain users’ attention. This design decision is in turn a result of the existence of a marketplace for attention. This giant marketplace (estimated to have a spending of $389.29 billion in 2021) has been made possible because of Advertisement Technology (Adtech). Then, it is fair to claim that “The problem” with online social platforms is high engagement with low quality content which is a negative externality of the Adtech Marketplace. This means that a possible approach to solving the systemic problem could be to understand and regulate this Adtech marketplace.

Regulating a marketplace seems far more tractable and acceptable - we know Economics and I think it is more or less taken as granted by policymakers that the State has to intervene when there are market failures. So now we have a problem with a big scale and a tractable intervention. At least in this forum, I don’t see many posts about Ad Tech. So I am going to assume Neglectedness. This then makes this a decent problem atleast for an individual EA to pay attention to, if not EA organizations. One could also argue that many EAs actually use social platforms - I wouldn’t have found EA at all without social platforms. Also if (EA) charities are spending part of their money on digital advertising it might be important to analyze if this is a good idea.

How do ads work

Adtech is way more complicated than you might expect! To illustrate this point in his newsletter, Prateek Waghre links to a visualization made by Luma Partners for the display ads ecosystem. Quite rightly he writes, “This one makes my head spin”:

So what follows is a highly simplified story of how an ad gets displayed. If a publisher (aka content producer/seller) wants to make money by displaying ads, they create a slot for it on their webpage and sign up on a Supply Side Platform (SSP) with an advertising ID. One can typically find the advertising ID of a website in different SSPs by going to their ads.txt page. For instance, try https://www.nytimes.com/ads.txt and you will see the advertising IDs of the New York Times. Now, when the publisher’s webpage is opened, the SSP’s server talks to the Demand Side Platform’s (DSP) server telling it that “User X has opened the webpage. Here is all the data we have on this user”. On the DSP, several advertisers (aka Buyers/Marketers) have signed up to show their ads. In this platform they can select what kind of consumer they want to reach - like so and so age group, so and so gender etc.. They also mention the CPM (Cost per mille/thousand) they are ready to pay. Now, if the advertiser’s preference matches the profile of the user who has currently opened the webpage, they can participate in an auction along with other advertisers who are also eligible. The ad of this auction’s winner will get displayed on the webpage and the fact that it has been successfully displayed is also relayed back to the advertiser. Remember that it typically only takes a fraction of a second between the user opening the webpage till the auction’s end. So as a simple picture:

Publisher’s webpage > SSP server > DSP server > Auction > DSP server > SSP server > Publisher’s webpage > User’s eyeball > Engagement data back to Advertiser

Advertisers don’t know what is going on

The biggest revelation for me was that this system is actually bad for the advertiser! Before I understood the problems of this system, I thought the advertisers were intentionally letting ads be against anything and everything - be it an extremist website or a cat pic - just in case they might get a click. But the truth is they have no idea what is happening! There is so much opacity in this system.

Advertisers actually have very little control of their ad inventories. They think they are buying attention with it. But what they are actually buying is a package that promises attention but it may not actually deliver on that promise. To quote Tim Hwang’s Subprime Attention Crisis (Ch4), “All an advertiser wins in an ad exchange auction is the right to display its content on a loading web page. When a demand-side platform (DSP) is programmed to seek out opportunities to reach a demographic like “males 18 to 24 living in the United States,” it tells us whom the advertising will ideally reach, but not whether the people who actually see the ad will be persuaded, or even interested” He goes on to say that advertisers are buying a derivative (the packaged inventory) and not the asset (the attention).

Put yourself in the shoes of a marketer. You could buy an ad and the system would report back saying you got an impression. But that ad might be at the bottom of the page where no one actually saw it or the user might have an adblocker but it still got counted as an impression and you still paid for it (although it is not entirely clear if this happens or not). In addition (especially if your DSP is an Ad Network), you probably don’t fully know what websites the DSP may have in their inventory (see this story and many others covered in the Branded newsletter by Claire Atkin and Nandini Jammi) - it could be an extremist website and then someone like Sleeping Giants could then hold you accountable. There are so many middlemen in the system. As an advertiser, you are paying the markup for each of these middlemen instead of paying just for the ad space alone. In addition to all this, there is rampant ad fraud. Malevolent Publishers have manipulated their ads.txt(see this story here and here), spoofed domains, used click farms and bots among other shenanigans.

This is not something negligible that happens in the fringe. This is the norm of this marketplace. In fact, Google has confessed that 56% of their ads are not seen. There are reports that suggest ad fraud takes $1 for every $3 spent on digital ads. This is also not happening to some marketers who are not experienced enough. Uber realized it had wasted $100 million on useless ads (see here and here). Chase bank reduced to advertising on just 5000 websites from 400000 websites and saw no change in the results [story here]. There is a legitimate opacity in this market.

Digital ads are overvalued

Because this marketplace is so bad, one would expect that prices would go down since marketers would ideally be paying less for such a low quality product. But this is not happening. The prices of digital ads are going up! More marketers are spending more money while neither the growth of users nor the amount of attention they can give is increasing in the platforms proportionally. So marketers are overvaluing digital ads. Maybe marketers assume that because they get so much data on impressions, digital advertising is better. But the target of marketing must be conversions or at the very least building awareness. It can not be just getting some data!

This overvaluation leads some to believe that there is an Adtech bubble. The book Subprime attention crisis by Tim Hwang goes so far as to claim that this scenario is similar to the Subprime Mortgage Crisis of 2008 when the housing bubble burst. He compares the ad inventories advertisers buy to the Collateralized Debt Obligations (CDOs) investors used to buy since there is so much opacity in both of them. Just as rating agencies gave CDOs triple AAA ratings and overvalued them, DSPs and SSPs are also overvaluing the attention an ad inventory can really get the advertiser. Just as mortgage issuers passed up the risk of a mortgage up the chain, DSPs/SSPs pass up the risk of owning bad inventory to other middlemen and assume no accountability. Given these parallels, Tim Hwang argues that the marketing bubble will burst and others seem to agree (see here). Interestingly, he takes the position that it is better for this bubble to undergo a controlled burst instead of preventing it from bursting. He writes, “Rather than trying to fix a broken market, we should work toward a controlled demolition that reduces its influence in the long run”. Given the opacity, the obvious policy recommendation is for more transparency. Tim Hwang also argues for it in his book and I have seen a policy paper by ASD with a similar proposal.

Concluding remarks

So these are my initial learnings. The most striking part to me is that just as users of these platforms don’t know how content is served to them, advertisers don’t know where their ads are landing. The platforms are unfair to everyone involved - users as well as advertisers. Except for the platforms and other middlemen, there are no real winners here.

There is another interesting quirk to ponder. One of the points I made early on was that regulating the Adtech marketplace would in turn address other issues like Information Disorder and Data privacy since platforms would have to redesign. But what guarantees that a redesign of the platform would actually change people’s behaviours? Remember that none of us have experienced a regulated internet yet. What if even after a redesign we continue the same behaviours because we simply don’t know how else to behave on the internet? I don’t know the answer to this. But it may be a good idea to pursue the regulation of the Adtech marketplace as an end unto itself and not just as a means for other ends. It feels problematic to let an unstable economic model that thrives on fraud & opacity, hold the internet hostage. That is enough reason to regulate it.

Meta