Table of Contents
This week, digital privacy activist group noyb filed a complaint against Meta with the Austrian Data Protection Authority (DPA). The complaint addresses Meta’s recent introduction of a “pay-or-okay” monetization model in the EU and EEA regions, as well as in Switzerland, for its Facebook and Instagram products. This model requires users to either consent to data processing for personalized advertising or to pay for a subscription to an ad-free version of these services to access Facebook, Instagram, or both (offered in a bundle). I previously examined this development when it was announced in Meta, subscriptions, and the EU’s Privacy Gordian Knot.
noyb’s complaint — the machine-translated text of which can be found here — articulates several perceived, alleged violations of the GDPR of Meta’s new subscription model. The complaint is expansive and includes numerous legal arguments, which I don’t feel qualified to assess. But to my mind, the most interesting of these accusations is that the price point that Meta charges for its subscription does not represent an “appropriate fee.”
The use of this term is deliberate: back in July, the Court of Justice of the European Union (CJEU), in a judgment on a case brought against Meta by Germany’s competition authority (Bundeskartellamt), issued commentary related to data processing for personalized advertising. The court argued that a social media platform cannot bundle the processing of the data used for ad personalization together with the data used to deliver the core social media use case under the GDPR’s contractual necessity basis, and it also called into question the use of the GDPR’s legitimate interest basis for this purpose (for more background on this topic, see this piece).
But the court also stipulated that a social media platform may offer users an alternative version of its product that doesn’t process data for the purposes of advertising personalization “for an appropriate fee” when a user does not wish to consent to data processing. From the judgment (emphasis mine):
Thus, those users must be free to refuse individually, in the context of the contractual process, to give their consent to particular data processing operations not necessary for the performance of the contract, without being obliged to refrain entirely from using the service offered by the online social network operator, which means that those users are to be offered, if necessary for an appropriate fee, an equivalent alternative not accompanied by such data processing operations.
This, in essence, is the “pay-or-okay” model, as I outline in more detail in Meta’s inevitable detour with EU privacy: “pay-or-okay”. noyb’s complaint dedicates significant attention to Meta’s subscription price, arguing that it does not meet the “appropriate fee” standard as articulated by the CJEU.
noyb is a well-respected and influential organization, and I commend its dogged and fervent defense of consumer privacy. But I believe that the parameters proposed by noyb in its complaint to test the “appropriate fee” standard for Meta’s subscription product, and in general for “pay-or-okay” models, are logically deficient.
To demonstrate this, it’s important to first identify the groups of users that are impacted by any social media platform’s introduction of an ad-free, “pay-or-okay” subscription product. The diagram below segments consumers into different groups, categorized by various dimensions:
- Their current status as a user of some specific social media platform that will or has introduced a subscription product;
- Their willingness to pay for an ad-free, “pay-or-okay” subscription version of that product;
- Their privacy sensitivity.
This is a rough mental model, and consumer receptivity to an ad-free social media product will obviously invite greater nuance.
Per the diagram, groups 1 and 5 will pay for the subscription: these are consumers — be they existing users or not — that are both privacy sensitive and exhibit a willingness to pay for a service that does not process their data for the purposes of advertising personalization. Group 1 can be considered to represent existing users who would prefer but do not demand the ability to use the social media platform without their data being processed for personalized advertising. Group 5 can be considered the consumers who refuse to use a social media service unless their data is not processed for advertising personalization.
In this model of consumer logic, group 6 represents the only group of consumers that will neither pay for a subscription nor consent to having their data processed for advertising personalization. But note that this group of consumers are not existing users: these are consumers who refuse to use a service that processes their data for advertising personalization but who also have no willingness to pay for a service that explicitly doesn’t.
Note that Meta’s implementation of the subscription poses a complication to the above logic, which is that the subscription eliminates all ads and not just personalized ads. This means that privacy sensitivity isn’t the exclusive motivator in adopting the subscription: consumers may elect to pay for the subscription simply because they want a completely ad-free product experience. This diversity of motivations means that groups 3 and 7 — consumers who are not privacy sensitive but are willing to pay for a subscription — may choose to pay simply to avoid ads.
noyb’s complaint argues that, while the CJEU’s definition of an “appropriate fee” is vague, the price point of Meta’s subscription product does not satisfy it. noyb proposes the following standards for determining an “appropriate fee” for a subscription service in a “pay-or-okay” model:
I question the prudence and logical rigor of these tests as articulated in the complaint, and I address them individually below.
A price that does not significantly manipulate the actual consent rates and thus the will of the user (see above 4.3.1).
In section 4.3.1, noyb cites the very high consent rates in some “pay-or-okay” implementations as support for the notion that this model undermines consumer agency. From the complaint:
A consent rate of around 99% speaks in itself against any free decision by users. Neutral surveys have demonstrated that only 3-10% of all users (taking an extremely generous view) want their personal data to be processed for personalised advertising on Facebook. In light of this fact, there is a discrepancy of over 90% to the actual inner will of the users.
This statement conflates freedom of choice with idealized consumer preferences. That “3-10% of all users…want their data to be processed for personalised advertising on Facebook” is not indicative of the tradeoffs that consumers are willing to accept in order to access a service for free.
Put simply: the 3-10% figure is irrelevant in this context unless the respondents in the poll were offered the secondary choice of paying for the social media product. “Would you prefer that some social media property not process your data to deliver personalized advertising?” is a fundamentally distinct proposition from, “Would you prefer to pay €9.99 or €12.99 to access some social media property, or would you prefer to consent to your data being processed for personalized advertising?”. This is a genuine expression of a tradeoff; a consumer is free to make the choice that best suits their needs.
A price that is affordable for those affected even with a further expansion of “pay-or-okay” systems.
noyb notes:
Assuming an affordable fee, it is easy to calculate that any realistic amount is exceeded here: Google assumes that a person has an average of 35 apps installed on their mobile phones. If only the average installed apps were charged a similar fee, this would result in a “data protection fee” of €8,815.80 per year. This does not include fees for websites or offline consent. Even with an average EU gross income of €34,750, this would result in a “data protection tax” of 25.3% per working person. For a family with two children, this would amount to €35,263.20 per year just to keep the family’s mobile phones free from data processing for personalised advertising.
This statement is predicated on a number of tenuous assumptions.
First, that all of the apps on this theoretical, average person’s phone currently monetize through behavioral advertising and would thus need to (or could feasibly) implement a “pay-or-okay” business model. This is unlikely. The app economy is diversified: a meaningful proportion of the apps on this theoretical, average person’s phone could belong to categories as varied as mobility (ride-sharing, scooter rental), travel, games, banking, and productivity (email apps, work-related messaging apps). These app categories do not, by and large, monetize via personalized advertising. Additionally, some of these apps would likely be for streaming services (Netflix, Disney+, Spotify) that already charge a subscription fee.
Second, that consumers only keep the apps on their phones that they use habitually and would consider paying for if forced to choose between a subscription and consent to behavioral advertising.
And finally, that all other apps command the same pricing power as Meta’s. Extrapolating pricing and usage from apps as pervasive and profitable as Facebook and Instagram to all other apps on a user’s phone is likely to overstate the notional cumulative monthly subscription expense. Meta’s monthly fee for subscriptions purchased in-app is €12.99 (the monthly fee for subscriptions purchased on the web is €9.99), and €12.99 * 35 * 12 = €5,455.80. The €8,815.80 number proposed in the complaint is derived by using the price of both the Facebook and Instagram apps as a bundle, which is €20.99 per month. To arrive at the €8,815.80 total yearly cost, the €20.99, two-app bundle price is applied to every app on the theoretical, average person’s phone. This also assumes that consumers won’t prioritize their payments or set a maximum budget on monthly subscriptions.
A price that covers the costs of the person responsible plus an appropriate mark-up (“fair income”).
The complaint calls into question the degree of margin that a social network should be allowed to extract for the provision of its services:
The following would apply when calculating appropriate remuneration to cover costs: The cost of providing Instagram and Facebook is around 18% of revenue. The cost of providing “social” networks is not particularly high; Mastodon, for example, is operated by non-profit organisations. Even with a generous profit mark-up, a “reasonable fee” would therefore be far below the €251.88 per year demanded by the complainant.
I find the focus on profit margins here to be misguided.
Exclusively highlighting Meta’s cost of revenue when considering margin is misleading. The cost of revenue metric ignores the research and development cost center, which captures new feature development, including machine-learning tools for advertising targeting and measurement functionality, as well as the sales and marketing and general and administrative cost centers. Per its Q3 2023 10-Q, the operating margin for Meta’s family of apps business unit was 52%, not 82%.
This standard doesn’t clarify anything and merely shifts ambiguity from the “reasonableness” of the product price point to the “reasonableness” of the margin achieved by the company. What is a reasonable profit margin? And who or what entity is entitled to determine that? Instituting restrictions on the profit that a company may generate — broadly, and absent any competition concerns — presents a slippery slope.
Mastodon’s user base is significantly smaller than Meta’s. Expense ratios simply aren’t comparable between Mastodon and mass-market social media platforms like Meta, TikTok, Snapchat, Pinterest, etc.: a non-profit could not operate a service at the scale of these platforms, given the absolute cost of the expertise needed to develop and manage them. Earlier this month, Mastodon revealed that its network had reached an all-time high of 655k active users (it’s unclear whether this relates to daily or monthly active users). In its Q3 2023 earnings report, Meta revealed that its “family” of products saw 3.14BN daily active people, on average, over the quarter.
A price that compensates for the profit to be made through personalised advertising, even if the respondent makes exorbitant profits here.
This standard posits that an appropriate fee for a “pay-or-okay” subscription product might be calculated as whatever price point allows the service to maintain its absolute amount of profit. noyb makes three arguments:
Firstly, the respondent forgoes the opportunity to place other types of advertising, such as context-based advertising. These losses are a decision of the respondent and therefore cannot be attributed to the complainant. According to a widely cited study, the difference in revenue between personalised and non-personalised advertising can be as little as 4%.
This is true: Meta’s “pay-or-okay” subscription product eliminates all ads, not just personalized ads. Were Meta to include, for instance, brand ads in its subscription product, the revenue loss relative to its ads product would be less severe and therefore the price point on the subscription needed to achieve profit neutrality would be lower.
But I believe the 4% benefit cited here is delusive and somewhat disingenuous, for two reasons. First, the academic paper that produced the 4% number studied the economic impact to publishers on cookie-based targeting. Social media platforms feature tools for targeting that are more sophisticated than cookies, given the closed-loop nature of the platforms, and they are able to integrate more extensively with advertisers’ products through pixels and in-app SDKs.
Second, the paper cited is far from the only paper that has interrogated this topic. Academic Garrett Johnson at Boston University has compiled a list of eight such studies, which produced estimates of the economic impact of targeted advertising of between 4% (the study cited by noyb’s complaint) and over 66%.
noyb’s second argument on this point is as follows:
Secondly, even if one (wrongly) assumes that all advertising must always be personalised, it should be noted that the respondent states an average turnover of $16.79 per quarter and user between Q3 2022 and Q3 2023 in Europe. At an Austrian National Bank reference rate of 1.0683 on 10 November 2023, this results in a turnover of € 5.24 per month and user. Even after deducting 20% VAT and levies to Apple of around 30%, the respondent would earn € 11.76 with the subscription and thus charges a surcharge of around 124.4% for refusing “consent” and using the subscription version. In this logic, he would suffer an economic “disadvantage” within the meaning of Recital 42 GDPR and there would be no voluntary consent.
As I point out in The EU’s Post-Advertising Internet, which unpacks the economics of a potential “pay-or-okay” subscription model and was published before Meta officially announced its product, Meta’s “Europe” reporting region includes Turkey, Russia, and the United Kingdom, none of which are eligible for the “pay-or-okay” model. Thus, the “Europe” ARPU cited here — $16.79 as a DAU-weighted average between Q3 2022 and Q3 2023, inclusive, which backs out to €5.24 per month per the currency conversion rate used by noyb — is difficult to compare against the “pay-or-okay” subscription price, which is only currently offered to users in the EU, EEA, and Switzerland.
But even if the “Europe” region mapped perfectly to the geographies that are eligible for the “pay-or-okay” subscription product in an apples-to-apples comparison, invoking ARPU introduces another problem: the user base is not a monolith, and some users are more prized by advertisers than others. Most scaled social media platforms offer conversion optimization mechanisms that automatically improve advertisers’ campaign targeting over time as the platform determines which users are most likely to convert (purchase, or otherwise engage with the advertised product) based on their shared characteristics and prior behavioral patterns. This is the “behavioral” in behavioral advertising. Users with an observed stronger affinity for certain types of products are targeted more readily by platforms than those without; not every user is valued at the same price by advertisers.
Revisiting the diagram above: the users in groups 1, 3, 5, and 7 likely feature higher ARPUs than those in groups 2, 4, 6, and 8, simply as a function of their demonstrated ability to pay for a subscription product. Removing these users from the group exposed to advertising will almost certainly reduce the overall ARPU of the remaining group to which ads are exposed. Thus, simply comparing the current ARPU to the subscription price and acknowledging a price delta does not take into account that would-be subscribers may feature an ads ARPU that is higher than the subscription price, even if the overall ads ARPU is lower.
To achieve a legitimate, defensible comparison, one must weigh the subscription price to the post-subscription blended ARPU across subscribers and free users. Given that the ARPU of ad-supported users is likely to decrease once subscribers decamp to the ad-free product, the subscription price would need to be greater than the current ARPU value to compensate. These are two fundamentally distinct analytical exercises.
noyb submits an additional argument related to profit compensation:
The flat-rate subscription prices also do not take into account the actual use and the actual turnover generated by the complainant on Facebook and Instagram (with “consent”). Especially for occasional users like the complainant, a flat-rate subscription model is unlikely to be economically appropriate, because a person who is online for 5-10 minutes per day pays €251.88 per year, just like a person who consumes Instagram reels for hours every day. Even if the “reasonable fee” were to compensate for the respondent’s loss of profit, this fee would have to be measured individually against the corresponding income resulting from its use. This is not the case here.
I find this argument curious. A metered monetization mechanic is, by definition, not a subscription. Further, the users that consume the most content on social media apps are not necessarily the most valuable to those apps — in fact, social media consumption might be inversely correlated with perceived value by advertisers, given its potential suitability as a proxy for productivity.
What is an “appropriate fee” for a social media subscription?
Without knowing the current ARPU values of the users in the EU, EEA, and Switzerland that fit into groups 1, 3, 5, and 7 above, it’s not feasible to derive a specific monetary value for an “appropriate fee.” But I would propose two alternative, conjunctive conditions for calculating whether a “pay-or-okay” price point represents an “appropriate fee”:
- The price achieves, at most, overall ARPU parity between the pre-subscription and post-subscription periods, and;
- The fee doesn’t materially exceed those charged by comparable services.
The first test is mostly distilled from my argument above about the segmentation of ARPU values for users that will subscribe and users that will not subscribe to the “pay-or-okay” product. Again, because the ARPU of the free product is likely to decrease as a result of an exodus of subscribers — who possess a demonstrable ability to expend disposable income — then the price point of the subscription product would need to be higher than the ARPU of the current ads-only product to offset that difference. An appropriate price point might be whatever achieves ARPU neutrality through the introduction and transition to a “pay-or-okay” subscription product: the price point that results in a blended ARPU equivalent to the ads-only ARPU calculated prior to the introduction of the “pay-or-okay” subscription product.
The second test is more specific: social media platforms don’t exclusively compete with each other for consumer attention but rather with all consumer apps. The prices of non-social media subscription products are instructive when considering an appropriate fee for a social media platform’s “pay-or-okay” subscription product.
This is especially true as many social media platforms (YouTube, Facebook, Instagram, Snapchat) shift the mix of their content formats to video. A selection of subscription products in Germany and their price points is presented below:
- Netflix: €12.99 / month for the Standard tier, 17.99€ / month for the Premium tier;
- Disney+: €8.99 / month for the Standard tier;
- YouTube Premium: €12.99 / month;
- Spotify: €10.99 / month;
- Amazon Prime (includes Prime Video): €8.99 / month;
- Apple TV+: €9.99 / month.
Viewed through this lens, the price points for Meta’s “pay-or-okay” subscription are not out of the range of market benchmarks. These prices are shaped by consumer demand. Subscription services are incentivized to optimize their pricing for total revenue and not to simply charge as much as possible. Meta is no different; charging an exorbitant price that isn’t supported by demand would undercut its financial outcome.
One might argue, however, that Meta’s “pay-or-okay” subscription product stands in contrast to these (aside from YouTube) because it is an alternative to a free version that is gated by consent and monetized by ads. In other words, some might contend that the problem with the price point isn’t that it exists, but rather that it exists as the only consumer access alternative to data processing for personalized advertising. This argument could be used to determine that, in the case where a subscription serves as a foil to personalized advertising, the price point cannot be anchored to any commercial rubric but instead should serve to provide the greatest possible consumer accessibility.
This is more of a philosophical than an economic position. And with respect to GDPR compliance, this issue has seemingly been addressed through precedent: newspapers like Der Spiegel and Bild utilize “pay-or-okay” models in implementations that have already been scrutinized by European DPAs. My contention is that, if an “appropriate fee” for a “pay-or-okay” subscription model can be determined, the locus of that pricing logic should be dictated by market benchmarks and ARPU neutrality.
