Copyright: tashatuvango

Digital advertising promises advertisers the holy grail of spending their advertising budget only on ads that are relevant for the consumers who see or hear the ad. Gone are the days when advertisers knew that half of their budget was wasted but did not know which half.

However, the digital advertising ecosystem also has produced consumers who are addicted to free services and give up their privacy to use them. It has enabled political manipulation and micro-targeted propaganda. It has created two digital advertising behemoths, Google and Meta, who are accused of unfair behavior towards publishers and advertisers. And there are growing doubts that digital advertising really delivers targeted ads to those for whom they are relevant.

So, what’s the problem? Where does the digital advertising system go wrong? What can be done about it? In this blog I analyze the problems and explain the solutions.

But first, a reminder: As explained at length by the Investopedia, There Ain’t No Such Thing As A Free Lunch. Here is a shorter explanation.

The Free Lunch Theorem

Digital advertising is one way to fund the free use of digital services by consumers. Google Search, Facebook, weather apps, news apps and many other digital services are examples of this.

Another way to fund a free service is the freemium model:  basic services are free, and this is funded by the fee that premium users pay. LinkedIn is funded by premium subscriptions (and by ads).

Yet another way of providing free services is to find a sponsor who funds service use. Telegram Messenger is an example of this. Wikipedia is a free service funded by many small donors (and volunteer labor).

A variation of this is to base the service completely on volunteer labor, for example Mastodon servers.

The final way to provide a free service is to fund it from taxes. Most digital services of governments are funded this way. Combinations of these business models are possible too.

These different ways to provide a free service prove what I call the free lunch theorem: Your lunch is free if someone else pays for it, with money or with labor. In the case of ad-funded services, the advertisers pay. In this blog I unpack this business model, identify its problems, and discuss how these can be solved.

What is the business model of the digital advertising ecosystem?

As usual in these pages, we take a business modeling approach. We analyze the business model of ad-based digital services from four perspectives [1]:

• the value proposition offered by the service,
• the value hierarchy that shows who participate in providing the service and what they contribute,
• the revenue model that shows cash flows among these actors, and
• the delivery model that shows how the value proposition is delivered to the customer.

All of these are models of a value network of business entities that jointly provide value to consumers and advertisers. Each shows the network from one perspective.

Value propositions of ad-funded digital services

Ad-funded digital services have two value propositions, one for consumers and one for advertisers.

• The proposition for consumers is that they can use a digital service without paying, in exchange for data about their demographics, behavior, and interests.
• The proposition for advertisers is that they can show targeted ads to consumers with a matching profile, determined by their demographics, behavior, and interests.

The consumer value proposition may be an information service such as the news or weather forecast, or a connection service such as a marketplace which connects buyers and sellers, search which connects searchers and publishers of content, or a social network which connects readers and writers of posts.

Whatever the consumer value proposition, an ad-funded service connects advertisers with consumers.

The revenue model

The revenue model of ad-funded digital services is simple. While consuming the service, a user provides attention and data to the provider. In the same transaction, the provider sells an impression to one or more advertisers that matches the profile of the consumer. Advertisers pay for this and also provide data to the service provider. The following diagram shows the flows of value among consumers, service providers, and advertisers.

An impression is an event in which a consumer sees or hears an ad. By targeting the ad to consumers with the right profile at the right moment when consuming a service, the advertisers hope that the impression creates the user’s attention, that this leads to a click on the ad, and that this leads to a conversion (sale).  Clicks and conversions are measured and paid for. For simplicity we only show impressions here.

The names between square brackets in the above diagram indicate value objects transferred among parties. The diagram shows that ad-funded digital service providers match the advertisers’ need for impressions with the consumers’ need for a service.

Counting individual impressions is not possible in the offline world. Ad-funded offline services place ads on a page or at a physical location where users consume the service. Placing the ad and consuming the services are different economic transactions, as shown in the following revenue model of offline advertising. Individual impressions are not measured in this model.

Both offline and online advertising are targeted: They aim at a market segment with a particular profile. However, offline targeting aims at the audience of a media channel as a whole. It aims at the readership of a magazine, the public that passes an advertising column, the viewers of a particular TV program. By contrast, online targeting aims at individual consumers. This is often called microtargeting, behavioral targeting or personalized advertising, but to make its nature perfectly clear, I will call it individual targeting.

In individual targeting, advertisers or their representatives decide whether to show an ad to a consumer based on, among others, the profile of this particular consumer. Other information, such as the contents of the web page where the ad will appear, will also play a role.

The opposite of individual targeting is anonymous targeting, where the consumer profile is not used in the decision to show an ad to the consumer. Offline advertising is anonymous, as is search advertising, where ads appear based on search terms.

Value hierarchy of the ecosystem

Underneath the simple revenue model of ad-funded digital services there is a huge ecosystem of programmatic advertising services and data services. Let’s go through the following value hierarchy from the top down. The model shows which entities participate in the value network and what services they provide. It is organized in a set of layers called the ad stack.

At the top we see that ad-funded digital services provide services to consumers and advertisers.  There are two kinds of digital services, information services and platform services.

• Information services provide news, weather forecasts, music streaming, traffic information, etc. through web pages or apps. They may offer the option to subscribe to an ad-free version.
• Platforms provide marketplaces, search, and social networking to consumers. They are all ad-funded. Marketplaces are usually additionally funded by transaction fees.

Ad-funded information services and platform services provide digital services to consumers, and advertising to advertisers. Digital ads are shown or played in empty space or time, called inventory, which is part of a screen or a media stream, called property. Ad-funded digital service providers are property owners who offer impressions shown on their inventory for sale to advertisers.

Ad networks are companies that aggregate the inventory of information service providers. For example, web page publishers and app developers can join an ad network, which will then handle the targeted advertising for them and share ad revenue with the publisher or developer on whose property the ad appears.

Digital advertising is supported by so-called programmatic advertising services. Supply-side platforms (SSPs) handle advertising on behalf of owners of inventory and demand-side platforms (DSPs) handle advertising on behalf of advertisers. When inventory is downloaded to the device of a consumer, ad exchanges conduct auctions to place ads on inventory.

In addition to matching inventory to an ad by means of ad exchanges, premium inventory may be sold directly to advertisers (programmatic guaranteed), or it may be offered to an advertiser for a negotiated price (preferred deal). An example of premium inventory is ad space on a high-quality web page such as the New York Times. An example of low-quality inventory is inventory on made-for-advertising (MFA) web sites, which contain low-quality content set up just to sell impressions.

Another way to sell premium inventory is by joining a private marketplace, ad exchanges that can be joined only by invitation. Private marketplaces have less problems than open marketplaces with brand safety (showing an ad only against content that the brand wants to be associated with) and inventory quality (avoiding MFA sites and other junk spaces) than open marketplaces.

These variations of matching inventory to ads are all part of the programmatic service layer of the ad stack but in order to show the essential structure of the ad stack, the above diagram only shows the parties in the ad exchange mechanism.

At the lowest level of the ad stack a variety of data service providers collect consumer data online and offline, curate and package it, and sell it to entities higher in the ad stack who want to improve their consumer targeting. Data service providers are sometimes called data brokers.

The above ad stack is a simplification that shows the essence of the digital advertising value network. More elaborate models show the size of the network in terms of participating entities. Dominik Kosorin provides information about how the entities in the programmatic advertising layer work [2].

According to Business Insider, global digital ad spending in 2021 was $506 Bn, or 63% of global ad spending. This is expected to grow to $870 bn or 74% of global ad spending in 2027.

Ideally, we should be able to elaborate the revenue model for the entire network, by making all value flows visible as we did above for the top two-and-a-half layers only (advertisers, consumers, service providers). In practice, this is too complicated, but there are heroic attempts to map at least some of the cash flows in the network, that provide useful insights.

Delivery model

A delivery model shows how a value proposition is delivered to the customer. It contains design decisions about technology, coordination processes, and data formats and semantics. A recent study by PwC done for the Society of British Advertisers (ISBA) reveals that the delivery model is extremely complex, lacks transparency, and is still immature [3]. Around 51% of ad spend in the study reached publishers, the rest ended up in the programmatic layer. An average of 15 percentage points of that could not be attributed in this study.

Branch organizations like the Interactive Advertising Bureau (IAB) define standards for digital advertising, such as OpenRTB for the auctioning process on open marketplaces, but overall, the delivery model is opaque even for its participants.

Problems

In each of the four perspectives on the ad-funded business model there are problems. I list them the following diagram. The arrows mean that if you change one part of the business model this will impact other parts. None of these problems exist in isolation, and solving one of them changes what you can do with the others.

Let’s zoom in these problems and their possible solutions in turn.

Consumer surveillance

Ad-based digital services require consumer surveillance on a massive scale because individual targeting requires individual consumer profiles. Platforms and data service companies collect data about consumer demographics, consumer behavior, location, and identity [4].

Demographic data may include age, gender, life status (single, married, divorced, with children, retired etc.), wealth, income, financial behavior, and health. Behavioral data is used to identify interests (games, sports, political inclination, etc.) and intent (purchase intent, voting intent, intent to move, to cancel service, etc.). Location data can reveal activities, shopping behavior, brand preferences (e.g. loyal Starbucks customer) etc. Identity data includes cookies, email addresses, device IDs etc.

The problem with consumer surveillance is that individual ad targeting violates privacy laws. Digital service providers may circumvent this by asking the user permission to track their behavior and serve them individually targeted ads based on this. But consumer consent, if given, is uninformed, because consumers do not know what data is collected about them, what profile is derived from this, and that their profile is used for purposes unrelated to ad targeting. Let’s unpack this problem. First, privacy.

Privacy

When an SSP offers an individual impression for sale on an ad exchange, it offers inventory, which is part of property where content is displayed or played, and the profile of the consumer who is about to view/listen to this property with ads included. The DSPs that represent advertisers read the individual consumer profile of an impression to decide on their bid.

This violates the privacy standards of the EU General Data Protection Regulation (GDPR). The GDPR requires explicit consent for processing personal data. The request for consent must be easy to understand, must mention the purpose of data processing, and must mention the organizations relying on the consent. There must be no nudges in the direction of giving consent, and if the consumer refuses, there must be an alternative way to access the service. In other words, ad-funded business models that depend on individual targeting must have an alternative in place, such as anonymous targeting. Finally, the consent must be easy to withdraw.

None of this is true of individually targeted ad auctions.

After five years of litigation, this month (August 2023), Meta has promised to include a request for permission for individual targeting on its properties in its consent form. It will provide more details in September.

If we assume that Meta finds a formulation of its privacy policies that covers its individual targeting practice, that agrees with the GDPR, then this will surface another problem: uninformed consent.

Nonsolution: Informed consent

Current attempts to rein in the surveillance economy focus on the privacy problem. A recent policy briefing of the EU reviews current and possible future actions to regulate surveillance based on the GDPR. The goal seems to be to phrase informed consent in a way that gives citizens control over the data that is collected about them.

However, consumers readily click any consent form that stands between them and free use of a digital service. Consumers prefer not to be tracked, but if tracking is needed to use an ad-funded services for free, they will consent to it. This is hardly a fair exchange. Consumers must hand over their data in exchange for using a service in a take-it-or-leave-it offer.

The consent consumers give to get access to free services is completely uninformed. Studies estimate that it would take one to two months of full-time reading each year to read all privacy policies you come across in that year. And you would need a college degree to understand them. No consumer reads or fully understands the privacy policies he or she consents to.

And even if they would spend the time and educate themselves on the privacy policies, consumers cannot know what the effect of their consent is. Profiling data allows the prediction of personal attributes such as ethnicity, religious and political views, relationship status, sexual orientation, alcohol use, cigarette consumption, drug use, credit score, economic stability, plans to have a baby, plans to change jobs, and more [5]. (A “prediction” in this context is the estimation of the value of an unmeasured variable from known variables.)

Most consumers are not aware of the size of the profiles that digital service providers keep about them. (Facebook alone collects, buys or derives 52 000 data points about each user.) This allows advertisers to target individuals on unfair grounds. For example, commercial advertisers may discriminate their offers based on race or religion. Political parties may target individuals based on their political profile and send different voters contradictory messages. Governments may manipulate individuals in other countries to stir up unrest. When giving their uninformed consent, consumers do not agree to these consequences; they just want to use a service for free.

Nonsolution: Individual data pods

A radical way to regain control of surveillance is to treat data about a person as owned by that person, and not by the company that collects or buys the data. The proposal is to create an infrastructure of data pods in which people can store their own data. They can then control who can access this data, ask a fee for that, and also learn a lot about their own behavior from their own data pod. In this solution, companies have to ask consent, not consumers.

This is an attractive idea, but managing access to your data pod may turn out to be a full-time job beyond the capabilities of most us. Think again of the 52 000 data points that Facebook has on you. And that is just Facebook.

Lanier & Weyl foresee non-profit mediators who do this on behalf of consumers [6]. That presupposes an amount of volunteer behavior that I do not see happening on the scale necessary to make this idea work. And if these mediators would be profit-driven, they would have to convince a sufficient number of consumers and advertisers of the viability of a market for consumer data. There are no signs that this is happening  ( [7] page 415).

Moreover, I do not expect Google, Meta, Amazon, and all data brokers in the ad stack to store their consumer data in decentralized data pods owned by the subjects of the data. This would be like asking all mining companies in the world to hand over all the ore that they have mined, and can still mine, to consumer advocacy movements.

To make the idea even more unworkable, data about me is data about the people I live and work with. The people with whom I visit events, undertake projects, and with whom I am photographed will appear in my data pod. Should we share the revenue of that data? Should we jointly grant or withhold permission to use the data? The idea of data pods cannot work.

Funding the profiling business

The problem is bigger than privacy violations and lack of control over data about you. Fundamentally the problem is the business model of the ecosystem as a whole.

As shown in the ad stack, programmatic advertising uses the services of data brokers, companies who are in the business of consumer profiling. Acxiom, Analytics IQ, Connexity, Experian, Kantar Shopcom, Lotame Data Exchange, Merkle, Nielsen DMP data, Oracle Data Cloud, (BlueKai Marketplace), Visa Audiences, and many, many other companies in the data services layer of the ad stack collect online and offline data, integrate it, curate it, and sell it to SSPs, Exchanges, and DMPs in the programmatic advertising layer [4].

The companies in the data services layer construct consumer profiles that are not only interesting for marketers, but also for many other private and public entities. Data service providers have more clients than those in the ad stack alone. Here are a few examples.

The largest financial data broker in the US, Yodlee, sold poorly anonymized bank and credit card transaction data to investment firms and hedge funds. The data could be de-anonymized with relatively little effort.

In order to predict crime and identify potential suspects, and to identify public safety-related events at an early stage, the FBI has contracts with Dataminr, which monitors social media posts, Ventell,  which monitors cellphone location data, and with Palantir, which visualizes relationships based on social media data, license plate numbers, and other data. Other government uses of profiling companies are the IRS, which uses Ventell to enforce tax rules, and the US Immigration & Customs Enforcement service to enforce immigration laws.

Apparently, individual ad targeting funds a profiling industry that feeds not only the digital marketing sector, but also other companies and government organizations with individual profile data. Profile data may influence whether we are offered insurance, where police patrols, and who is considered a potential (!) suspect. Consumer surveillance turns into citizen surveillance.

The examples listed above are about US-based companies. But these companies collect data about digital services used globally and offer their services world-wide. Government agencies in Europe, the middle east or the far east can buy data from these companies.

This is a problem not only because it can have harmful consequences for individuals. It is also bad because this industry operates in stealth mode and avoids democratic accountability.  While it aims to make citizens transparent, it hides itself in obscurity. While its business is to watch people, it does not like to be watched.

Solution: Ban individual ad targeting

The problems with privacy violations and the surveillance business can be solved in one strike by banning individual ad targeting. This would eliminate trading personal data in ad exchanges or in human negotiations between publishers and advertisers. And this would remove the largest market of data brokers.

Several parties have advocated a ban on targeted advertising. For example, the European data protection supervisor (EDPS) has proposed including this in the Digital Services Act. This was watered down to a ban on individual targeting minors, and on using sensitive attributes such as race or sexual orientation in ad targeting. The watered-down version still allows trading personal data in the programmatic layer.

The Electronic Frontier Foundation has given guidelines for Congress to draft a bill to ban behavioral targeting. (The bill must still be drafted.) The British Competition and Markets Authority (CMA) proposes a slightly weaker intervention, namely to require individual ad targeting to be switched off by default, and provide consumers an option to switch it on [7].

The CMA proposal would still allow trading personal data on ad exchanges, which is still a privacy violation, even if a consumer clicked an uninformed consent button. So I side with those proposing a complete ban on individual ad targeting.

Banning individual ad targeting would remove the primary incentive for consumer surveillance. The ban would prohibit the use of individual profiles in auctioning off an impression, but it would allow targeting based on contextual criteria such as the contents of a web page or the keywords in a search query.

This would prevent advertisers to follow you around when you have shown interest in buying a new camera. It would also make it impossible for advertisers to follow high-value individuals after they visited a premium site, until they visit a cheap website and buy their impression on that site, a practice that drains ad revenue from premium sites.

Banning individual ad targeting would still allow digital service providers to offer you personalized services. Netflix could still construct your viewing profile and recommend you movies based on that. The ban would just eliminate individual ad targeting, not personalization of services.

This would not kill the advertising industry, but it would some excesses of the surveillance economy. The incentive of advertisers to see individual profiles would disappear. The conflict with the GDPR would be removed. The data providers at the bottom of the ad stack would lose a large market, which is fine.

Advertising revenue of digital service providers may decrease, because advertisers will pay less for anonymous impressions than for individually targeted impressions. Some services may have to supplement their ad revenue with revenue from subscriptions, either in combination, as newspapers do, or in a freemium model, as Spotify does. Newspapers show ads and ask subscriptions, Spotify offers the choice between a free ad-funded service and an ad-free paid service.

Consumer addiction

Engagement

Next to surveillance, consumer addiction is a problematic side-effect of some ad-funded digital services.

For some digital services, like search, it is a sign of quality that a consumer uses it briefly. The answer to a search query is best if the user does not continue searching after seeing it. If the search result does not satisfy the user, the user will continue searching by entering a new query. The less satisfying the answers, the longer the search session. Search engines want their sessions to be short.

But unlike search engines, ad-funded social network providers want each session to last as long as possible, in order to maximize the time in which users can be shown ads. They want their users to be “engaged” with the service. They tend to do this by promoting sensational or inflammatory content, because this glues users to the screen.  This has led to users being addicted to filter bubbles of hate speech, fake news, disinformation, and propaganda. Ad space on this content may be of low quality and sell at a low price, but if the network sells a lot of it, it makes for great revenue. The problem is that this has addictive and socially destabilizing effects.

What can we do to manage this?

Nonsolution: Banning individually targeted posts.

The mechanism by which the user’s screen time (“engagement”) is maximized is individual targeting of free, “organic” posts based on past behavior and other profile data. So why don’t we ban individually targeted posts, just as we proposed for individually targeted ads?

Banning individually targeted posts would imply banning personalized services. It means that we would not allow any digital service to use past consumer behavior on that service to personalize the service. But many forms of personalization are harmless. There is nothing wrong with Netflix recommending you a movie based on your past viewing behavior. Banning individually targeted posts is not a proportional solution to the problems of addiction and destabilization caused by social networks.

Nonsolution: Consumer data transparency

A less drastic idea is to make consumer profiles visible to consumers. Each of the users of a digital service should have the right to see the profile that a digital service provider has of them. The GDPR actually mandates this.

The idea is that if a consumer sees that data about them is incorrect, they can demand correction. And if they don’t want the provider to store this data about them, they can demand removal.

But just like individual data pods, this too goes way beyond the constraints in time and intellectual effort of nearly all social network users. For example, even though I opted out of all tracking by Google, requesting the profile data that Google has on me yielded a dump of 27Mb consisting of 118 files in Excel, html, json, and other formats. Without extra software support and time on my hands there is nothing I can do with this.

Instead of loading the consumer with data management obligations, it is more effective to offer the user a choice to move to competing service. This is a change in the business model of social networks: it would introduce competition among social network providers. The wat to make this work is to demand interoperability of social networks.

Nonsolution: Opting out of individually targeting of posts

Another less drastic idea is to offer consumers the option to opt out of individual post targeting. As an alternative, they could be presented with post chronologically or based on some other criterion. The DSA mandates this.

Including the possibility to opt out of personalized post targeting is the start of a solution. But we have to see how this works out at scale for the global user base of social networks. Do all users understand the option? How does the provider describe the different options?

A more convincing solution would be to mandate that consumers can switch with little effort to a competing social network provider, without losing their friends nor the ability to communicate with them. This requires network interoperability. I will discuss this option next.

Solution: Interoperability

If we require true interoperability among social networks, a consumer can switch to another network without losing their ability to communicate with their friends on their old network. With true network interoperability, social network providers would jointly offer a universal service, just like telco providers do, and compete with each other for customers in terms of price and quality, without keeping their customers captive. If customers do not like inflammatory content shown to them by a provider, they can move to another provider that does not show this content.

Competing providers could offer a subscribed, ad-free service, or they could offer an ad-funded free service. To attract customers, they could clarify what their matching algorithm is and optimize it towards an attribute valued by their customers, such as diversity of content, geographic location of the content provider, position of the content provider in a social network, or some other criterion. This would be similar to ad-funded newspapers with an editorial policy. There are many other competition options.

Solution: Amplification responsibility & ad repository

Even in an economy of competing and interoperable social networks, consumers may prefer to join a network that exposes them to hate speech, racism, misogyny, or other inflammatory content. And in the weaker solution mandated by the DSA, offering the user to opt out of individual post targeting, many users may opt to receive personalized junk. As explained by Gao Han, a senior UI designer at ByteDance and employee number 22, “We must face the fact that for 96% of the people, their needs are so vulgar” ( [8] page 85).

People are free to eat junk, and companies are free to offer it. Nevertheless, with freedom comes responsibility. Social networks in the USA currently enjoy the protection of section 230 of the Communications Decency Act, which says that networks may refuse to distribute content but are not responsible for the content that they do distribute. The result is a network of billions of users who can rely on a free broadcasting channel to transmit hate speech, misinformation, disinformation, propaganda, fake news and other sensational, inflammatory, or disgusting content to billions of users. Surely, the channel bears responsibility for amplifying these messages to keep their users glued to the screen?

Compare a social network to a post office who decides what mail you receive. It sends you photos and diary entries of your friends, messages of political parties, flat earth theories, conspiracy theories about the moon landings, racist content, fake news, and information about gardening. It streams cat videos as well as videos of live shootings and suicides. Wouldn’t you take the post office to account for this? Social networks did not author these messages, but they did read them and decide to send them to you. They decided to send these messages, not others; and they decided to send them to you, not to others. And they do this on a massive scale. Social network providers should take responsibility for their decisions to amplify content.

Solution: Anonymous distribution register

To enforce responsibility, the amplification by a social network should be visible. There should be an anonymized register that shows how often a post is distributed by the network, based on which characteristics of the user and the content. The contents of the post may be invisible, but the keywords that motivated its distribution to a target group must be visible.

Consumers may browse this register, but more to the point, journalists, researchers, activists, and law enforcement can too. This should help social networks take responsibility for their prime activity: boosting messages to their users.

Conflicts of interest

Consumer surveillance and consumer addiction are problems in the value proposition offered by ad-funded digital services. The next problem in the digital advertising ecosystem is a problem in the value hierarchy: The ad stack contains numerous conflicts of interest.

For example, Google Ad manager contains DoubleClick for Publishers, an SSP that represents publishers; it also contains ADX exchange, an ad exchange where publishers offer their inventory for sale. So the market master represents some of the sellers on the market. And it has something to sell as well, e.g. inventory on YouTube.

As another example, the Google Marketing platform is a DSP, representing advertisers.  Google itself also offers inventory. So the property owner represents advertisers who want to buy inventory on his and other properties.

In the colorful metaphors of Cory Doctorow,

It’s like a realtor representing the buyer and the seller, while buying and selling millions of homes for its own purposes, bidding against its buyers and also undercutting its sellers, in an opaque auction that only it can see.

It’s a single lawyer representing both parties in a divorce, while serving as judge in divorce court, while trying to match one of the divorcing parties on Tinder.

The proposed bipartisan AMERICA Act spells all possible conflicts of interests and prohibits them. According to the Act, a company with more than $20 bn in advertising revenue may not

• own a digital ad exchange if it also owns an SSP or DSP, or sells inventory itself;
• own an SSP if it also owns a DSP,
• own an SSP or DSP if it buys or sells inventory.

You can figure out how many conflicts of interests are exhibited in the ad stack shown above. The aim of the AMERICA Act is to create healthy competition in the digital advertising ecosystem by splitting companies that violate the Act along the lines of conflict.  In anti-trust law, this is called structural separation.

The British CMA has also proposed structural separation along the same lines ( [7] page 400). The EU has not proposed a law to this extent but is preparing to sue Google on its current antitrust laws on self-preferencing ADX by its SSP and DSP services.

These processes will not lead to any breakup of the ad-tech giants anytime soon, but in the long run the outcome should be the removal of the many conflicts of interests in the ad stack. This is good news for publishers, including news publishers, as it creates a fair playing field for publishers to generate revenue from ads.

Lost ad spending

Turning now to the revenue model of ad-funded digital services, there are two problems. First, lost ad spending. For several years, the advertising branch has struggled to figure out where in the ad stack their money ends up. A study of 1.3 bn impression on British web sites, commissioned in 2020 by the Incorporated Society of British Advertisers (ISBA), found that 15% of ad spend between the DSP and SSP of impressions could not be traced [3]. ISBA calls this the “lost delta”. In a follow-up study done two years later, this was reduced to 3% [9].

In 2023, the USA-based Association of National Advertisers (ANA) published a study that compared logs of three DSPs and six SSPs to analyze ad spending of 21 advertisers [10]. They found no lost delta, which is good news for the study participants.

The studies should be repeated on a larger scale so that advertisers can trust that no money is lost wherever they spend their ad dollars. The three studies investigated only a small part of the programmatic advertising layer. Most importantly, the walled gardens of the advertising giants Google and Meta have not been included in the studies, so that we know nothing of the destination of ad spend in those supply chains. Jointly, they command more than half of the digital advertising market but they constitute the most opaque parts ( [7] page 410).

The ISBA and ANA studies reported considerable difficulty unearthing the numbers, which points to a problem of lack of supply chain transparency, and also indicates a possible solution: standardizing data logs. Standardized data formats for logs should help researchers and auditors to understand and assess what happens in the digital advertising supply chain. The CMA proposes introducing a common impression ID to improve transparency about fees and bidding data ( [7] page 407).

Lost value

The ISBA and ANA studies also showed that participating advertisers appeared on an average of over 40 000 websites, most being non-premium. Apparently, they buy large numbers of low-value inventory. The ANA study estimated that 21% of impressions happen on Made-For-Advertising (MFA) web sites, which are sites containing clickbait and other junk made to generate ad impressions. This is 15% of ad spend.

Impressions on this long tail of web sites are cheap but also have low value for advertisers. Advertisers appear to prefer cost reduction (low price per impression) over value maximization (more revenue per impression). The study advises advertisers to buy inventory on private marketplaces, which contain premium inventory, or buy from publishers directly.

There is however a good reason why advertisers like remnant inventory. Individual ad targeting allows them to follow a high-valued customer, from a premium site such as the New York Times to a remnant site and buy an impression from that customer there. This is one way in which individual ad targeting drains income from premium publishers. At the ecosystem level, this is a race to the bottom.

To understand where ads appear and why, advertisers and verification service providers would be helped by an ad repository with the content of the ad, name of advertisers, targeting criteria and name of the web site or app where the ad appeared. The repository would also help consumer advocacy organizations to understand why consumers are targeted, and why. In fact, the EU DSA (article 30) mandates such a repository specifically for consumer protection. Here, I argue that it would be just as useful for the protection of advertisers.

Inaccuracy

In the delivery model, the core problem is whether targeted ads are actually delivered to the targeted group. This falls apart into a few questions: Are the consumer profiles accurate descriptions of the consumers? Are the content descriptions of web sites accurate descriptions of the sites? Are targeted ads actually delivered at the specified target?

The first two questions are about the accuracy of descriptions: Do they match the real world? The third question is about who wins the competition to be delivered at the desired target.

These questions are hard to investigate. Ideally, one would want independent verification providers to investigate them. But the advertising giants Google and Facebook restrict access of verification providers to underlying data ( [7] page 410). For those intermediaries where researchers could get access, such as in the ISBA and ANA studies mentioned earlier, access was very hard to get, took more than a year, and concerned only a fraction of auctioned ads.

Circumstantial evidence suggests that accuracy is low. In 2020 Procter & Gamble slashed $200M from its digital ad spend budget but yet reported a 10% increase in reach. After the GDPR was introduced, the New York Times cut off ad exchanges in Europe and kept growing its ad revenue.

An academic study on the accuracy of individual targeting, based on consumer profiles, reported that the cost of individually targeted standard display ads is 151% of the cost of untargeted, anonymously targeted ads, but that individually targeted ads have an improved hit rate of 121% [11]. For these ads, individual targeting was not cost-effective. For more expensive video ads, the improvement in accuracy was better than the increase in cost.

Accuracy of delivery depends on the attribute selected on. Targeting on gender was slightly worse than random targeting, but targeting on interests (sports, fitness travel) showed a slightly better hit rate than random targeting. Note that an even more interesting comparison would have been to compare individually targeted with contextually targeted ads. This was not done in the study.

More research needs to be done on the accuracy of targeting. But this study suggests that individual targeting may not always be accurate or cost-effective. Individual targeting resembles spam, because it targets an audience rather than the show they are watching. They don’t relate to what the consumer is doing and are therefore ignored.  Interest-based targeting, e.g. based on search terms or the contents of a web page, is more effective because relates to what the consumer is doing at the moment he or she sees the ad.

Evidence against the accuracy of individual targeting has mounted to the level that some authors now call advertising the time bomb at the heart of the free internet, comparable to the subprime mortgage crisis in banking [12]. Tim Hwang expects that once advertisers learn about the junk they are buying, they will withdraw from programmatic advertising on remnant inventory.

Without further data, it is impossible to say how (in)accurate targeting is. We need more independent verification providers and academic researchers who answer the research questions listed above. If we ignore individual targeting (which should be banned anyway), then researchers and verifiers must be able to see which website the ad was served, what other content was on the site, whether it appeared in the screen, how much of it appeared on the screen, and in case of music or video streams,  whether it was played, how long it was played, whether sound was on etc. Collecting and analyzing this data is compliant with the GDPR.

Summary

Free services are paid for by someone else than the user. They can be funded by ads, by premium subscriptions, by sponsors who give donations, by volunteer labor, or by taxes. Ad-funded services offer a free service to consumers and sell consumer attention to advertisers.

All advertising is targeted. Targeting can be based on contextual criteria such as the contents of a web page or search terms, date and time of the impression, or location of the impression to be sold. And it can be based on the individual profile of the consumer whose impression is sold. In anonymous targeting, no consumer profile is used.

The business model of ad-funded digital services contains a number of problems. First, the value proposition of individual ad targeting requires extensive consumer surveillance. Second, funding by ads motivates social network providers to keep consumers glued to the screen as long as possible. I reviewed four possible solutions that I think will solve these problems.

Consumer surveillance violates privacy rights, because individual consumer profiles are sold as part of impressions to be sold to advertisers. Solutions involving consent have not worked so far, and there is good reason to believe they will never work. A solution guaranteed to work is to ban individual targeting. This would take away the motive for consumer surveillance for publishers and advertisers.

Banning individual ad targeting would still allow anonymous targeting based on context and content. Ad-funded social networks would still have the motive to keep their users glued to the screen as long as possible, and an important mechanism to achieve this is to serve them with sensational, inflammatory, or disgusting content. This has addictive and socially destabilizing consequences.

For social networks, personalization of services includes individual targeting of organic, unpaid posts.  It is disproportional to prohibit personalization of services, but there are other ways to reduce the addiction problem for social networks.

• Interoperability would introduce competition among social networks, which would include competition on the quality of their matching algorithm.
• For networks which attract users by offering sensational and inflammatory content, the protection of Section 230 of the CDA should be replaced by responsibility for the choice to promote and amplify posts based on the content of the post and the profile of the user.
• Amplification of messages by social networks should be made transparent by a register of distributions that shows how often a post is distributed by the network, based on which characteristics of the post and the user.

The root cause of the two problems with the value proposition of ad-funded digital services is that consumers play two opposing roles in the ecosystem: As beneficiary of a service, and as resource for data that enriches an impression. In the advertising ecosystem, the role of the consumer is to be a source of profile data, which makes impressions more valuable, which generates more revenue for the key players in the ecosystem.

The other problems reviewed in this blog are about the structure of the ecosystem, the flow of value in the ecosystem, and about the delivery of value to customers (advertisers).

The programmatic advertising system is a duopoly of Google and Facebook, which occupy several roles in the ad stack that have conflicting interests. This is to the detriment of publishers, who get less revenue, advertisers, who pay too much for impressions, and consumers, to which higher ad prices are passed on. The solution to this is to structurally separate these advertising giants into parts that do not harbor these conflicts.

The flow of value in the ecosystem, and the delivery of value to advertisers, has a transparency problem. Money is lost between DSPs and SSPs, remnant inventory sold in auctions has no value for advertisers, and it is not known whether ads are actually delivered where they should be. The solutions to this are to create an ad repository that provides clarity about which ads are delivered where, standardize data so that it can be aggregated and analyzed, and allow independent measurement entities access to the data.

I also reviewed a list of solutions that cannot work: uninformed consent, individual data pods, banning individual targeting of organic posts, and data transparency for the consumer. We are not in the position to redesign the entire internet for an audience of well-informed, rational consumers who have the time on their hands to enforce all their rights.

The measures reviewed here require government intervention and regulation. We need the visible hand of government to give room to the invisible hands of democracy and the market to the digital advertising ecosystem.

References

[1]	R. Wieringa and J. Gordijn, Digital Business Ecosystems. How to Create, Deliver, and Capture Value in Business Networks, TVE Press, 2023.
[2]	D. Kosorin, Introduction to Programmatic Advertising, AdTechResearch.com, 2016.
[3]	PwC, “Programmatic Supply Chain Transparency Study,” ISBA, 2020.
[4]	D. Kosorin, Data in Digital Advertising. Understand the Data Landscape and Design a Winning Strategy, AdTechResearch, 2018.
[5]	W. Christl, K. Kopp and P. U. Riechert, “Corporate Surveillance in Everyday Life,” Cracked Labs. Institute for Critical Digital Culture, Vienna, 2017.
[6]	J. Lanier and E. Glen Weyl, “A Blueprint for a Better Digital Society,” Harvard Business Review, 26 September 2018.
[7]	Competition and Markets Authority, “Online Platforms and Digital Advertising. Market Study Final Report,” CMA, 2020.
[8]	M. Brennan, Attention Factory. The Story of TikTok & China’s ByteDance, www.chinachannel.co, 2020.
[9]	PwC, “Programmatic Supply Chain Transparency Study II Summary,” ISBA, 2023.
[10]	PwC, “Programmatic Media Supply Chain Transparency Study: First Look,” ANA, 2023.
[11]	N. Neumann, C. E. Tucker and T. Whitfield, “How Effective Is Third-Party Consumer Profiling and Auddience delivery? Evidence from Field Studies,” Marketing Science – Frontiers, no. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3203131, 2019.
[12]	T. Hwang, Subprime Attention Crisis. Advertising and the Time Bomb at the Heart of the Internet, FSG Originals x Logic, 2020.