Good morning,
Thanks to the absolute deluge of news over the last couple of weeks I was not able to dive as much into earnings as I would like to; that’s particularly unfortunate this quarter because this is one of the most momentous quarters for earnings in a long time, as it is the quarter where Apple’s App Tracking Transparency changes are really having an effect.
ATT is obviously an issue that I have been tracking heavily since Apple announced it at WWDC 2020, but Eric Seufert was way ahead of me: he predicted it on his blog Mobile Dev Memo back in 2017. I first interviewed Eric last February before ATT was rolled out with iOS 14.5 about what was coming; now that it is here I wanted to catch up with him to understand what happened, and what’s next.
I have to warn you: this interview ended up going very long, and got very geeky. I personally found it absolutely fascinating about not only the future of ads — and by extension the future of mobile gaming and e-commerce — but also the competitive position of Facebook and Apple in particular; still, this may be one you want to listen to via podcast.
On to the interview:
An Interview with Eric Seufert about the Impact of ATT
This transcript has been lightly edited for clarity.
Eric, we first talked I think it was actually earlier this year, but given the way that the ad market has shifted, particularly in the last six months, it’s a very tangible before and after feeling where we were on here at the beginning of the year — talking about this ATT change is coming, it’s going to have a huge impact, playing out which companies would be impacted more, which would be impacted less, what was Apple’s motivations in doing this — and now it’s happened, and so I wanted to get you back on, particularly after this quarter’s | results to pick up the pieces, see what we got right, see what we got wrong and what lessons there are to think about going forward. So thank you for jumping back on.
Eric Seufert: Thanks for having me. Yeah, this is like the dramatic before and after photo where the guy’s pants are like twice as wide as his torso after the regimen.
Yeah, I guess it’s the ad networks that lost a lot of weight this quarter.
ES: Yeah, yeah. Quite the famine.
So it was really interesting because I think the first results that came down the pipe were Snap’s. I always feel bad for Snap that they had to go first, on the other hand, I kind of don’t feel for Snap because they were so dismissive of the impacts of these changes — I mean, what were they thinking? I think we both wrote about this at the second quarter, about how they seemed awfully blasé about the potential impacts of these changes. What is it that blindsided them to a degree where their stock went down by 25% the day after earnings? What did they miss? And was that a miss that was widespread? Or was that something that was unique to them?
ES: It was unique to the class of ad platform that they operate. So when you think about the mobile ads ecosystem, there are these separate categories of ad tech companies, or companies that operate ad platforms. The category that Snap is in, I call them ad platforms, they’re also called self-attributing networks, but basically a company that owns an app and they sell the owned inventory in that app. So they own the content, they own the inventory and they sell it directly to buyers.
Just to make sure I understand, these are companies where they gather their signal from other apps, but then they display all the ads that they show within their own app. Is that the two pieces of it?
ES: Well, they could. There are tiers within that category. So Snap and Facebook, they were the ones that were ahead on building those behavioral profiles where they harvested this data from third-party properties, like from their advertisers. Twitter’s in that category, but they didn’t do that. They didn’t actually build those kind of profiles in a meaningful way that advanced their capabilities as an ad platform and so for that reason, they were insulated this quarter, they had a decent quarter.
Yeah, a classic example of Twitter being bad at business actually worked to their favor.
ES: (laughing) Yeah, exactly. The public market’s reaction to Snap was so severe not because I think that they are impaired more by ATT, I actually think they’re probably impaired less than Facebook by ATT, but just because they had done such a poor job of managing expectations. Like you said, they were incredibly confident for the last two quarters, totally hand waving ATT away saying it was the have very little impact. Evan Spiegel was championing ATT, “This is a great change that Apple is instituting. It’s privacy forward and we support it.” The reality is that they were impacted severely, I think not as severely as Facebook, but nonetheless severely. In the run-up to this quarter, their stock was basically at an all time high. Now it has moved down, it’s still in the fifties today.
The way I was anticipating it, I think you and I were in the same boat here, is you look at companies and what percentage of the revenue is direct response advertising, where you look at an ad and you immediately act on it, brand advertising on the extreme other end where, “Maybe I’ll buy this deodorant ten years down the road because I’ve been inundated with Old Spice ads”. There’s obviously a vast spectrum between there but the degree to which a company fell on either side was the expectations for this. Facebook’s way on the direct response side, Snap is 60/40 towards the direct response side, Twitter’s way towards the brand side, because the brand in some respects is easier to sell because it’s also harder to measure, those things are sort of interrelated to each other.
I think the weird thing to me about the Snap thing, and I think you’re exactly right about expectations, is I felt like the results came, if you understood that spectrum, the results by and large matched that spectrum. Facebook did get hit the worst, but because they had been pretty aggressive in saying they were going to get hit — I actually think they got hit worse than they suggested, but people built in that doubt into the model — I mean, their stock was down 5% or something, but relatively speaking, they were also committing to spending billions of dollars on this metaverse thing. I thought it turned out pretty well for them, all things considered.
ES: It was up after hours but I mean Facebook threw the kitchen sink at this. First of all, Facebook was down 5% last quarter after earnings, when they just acknowledged that, “Hey, ATT’s going to create headwinds.” So they were down last quarter. Then they were down after Snap reported like 7% and then they were mixed after hours after earnings and then the next day they dropped 5%, so all combined, in aggregate, they were — I mean, they were at $380 not that long ago.
Yeah, in September I think.
ES: $320 today, right? They’re still up for the year, but nonetheless, the compounded effect of that is pretty dramatic. But they acknowledged it, right? They took the medicine last quarter and then they pulled a magic trick that they can only do once, which is the whole rebrand and the breaking out reporting. I think you’re exactly right about what that’s going to do. When they break out the Facebook family of apps and then Facebook Reality Labs is separate reporting entities, it’s just going to showcase how incredibly profitable the Facebook ads business is. But I mean, they made a whole spectacle out of this. I also do think the whistleblower stuff gave them the ability to talk about something else in the call, which just muddied the waters.
That is the ultimate contrarian take, where it actually gave them a better thing to talk about, whenever everyone would consider the opposite. I mean, it makes sense.
ES: They spent a lot of the call, I mean, Mark opened the call with that, and he got angry, that was the most animated I think I’d ever heard him. I think it just distracted from, “Hey, we’re going to spend $10 billion this year and more every subsequent year on Facebook Reality Labs. Q4 is going to be soft, Q3 was soft, revenue was down quarter over quarter.” I mean, those are pretty dramatic things to announce but in aggregate, it was like a mixed signal, you couldn’t really focus on any given one of those. I think the whistleblower stuff is something that’s like, it’s abstract, you can just talk about this abstract, febrile thing versus actually having to focus on, “Hey, quarter over quarter, we were down.” And I don’t know that’s ever happened Q2 to Q3 in Facebook history, at least not in the last few years.
Yeah, I said it was a contrarian take, that doesn’t mean it’s a wrong take, I actually think it’s a very good point.
One thing that was striking about Facebook’s comments is despite the fact that they’ve been working on this earlier and much more intensely than Snap has, there’s some degree where their business is so much more mature than Snap’s that it’s going to take them longer to recover from this. They were talking in terms of years, as far as gaining back this capability. What does a future advertising platform look like in this ATT world? Like, how is it going to work?
ES: You basically have to rebuild the foundation of the house. Everything was built around these IDFA index profiles and that’s just going away.
And just to be clear, can you talk through — and this is something that I tried to write about, but I think you can articulate it much better — the fact that targeting and attribution are kind of the same thing when it comes to this sort of marketing. How did it work and then how will it work? I think I’d like to understand better.
ES: So there’s a clear distinction between the way that Facebook and Snap operate in that regard, and also there was a clear distinction in how they articulated this in the call. So Sheryl [Sandberg] said in the call, “The accuracy of our ads targeting decreased so the measurement of those outcomes became more difficult.” So she acknowledged that yin and yang, the two parts of the feedback loop. There’s the measurement, which is just observing some conversion that happened as a result of an ad and then there’s the round trip that takes, back into the targeting model, enriching the user profile on the basis of that observed conversion, such that you can target better ads to that person in the future. You know more about what they like and so therefore you’re better able to show them relevant ads. And in the Snap call, I don’t remember who said it, but they said it’s mostly a measurement challenge, so the ability to observe those conversions was eradicated. They weren’t doing as much with that round trip though, I don’t think they were doing as much to like build these very robust user profiles, such that they could target better ads as a result of observing those conversions.
I want to ask about that because I noted too that they described it very differently. I interpreted that though as them both saying the same things just in different ways. But do you actually feel like Snap was actually not that good at building these detailed user profiles, and they were more simplistic about it? Which in some respects is a good thing, because it almost feels like that would make them be able to recover more quickly even though they seemed more behind the eight-ball here.
ES: I do think that’s probably the case. I mean, first of all, they have fewer users, they generate less data, their ads platform is younger. They rolled out some of these products years behind when Facebook did so I think just for those very straightforward reasons they probably had a less mature product and also Facebook’s just got a much bigger org around this. Also before the Snap exec said that, he said something like, “No, we think our ability to target is basically uninhibited relative to where it was before. It’s just that we’re not able to measure the outcomes and so, for that reason, our advertising suffered as a result of that.”
So the process — that round trip is very important. Facebook serves an ad to you, you click on the ad, you go to a website or you go to an app. In either one of those, the destination would happen from an ad in the mobile app, that’s where 95% of their ad revenue comes from their mobile products, Instagram or Facebook Blue. You click the ad, you go to either website or an app and then when you do things on those properties, Facebook observes the fact that you do those things, either through the pixel on the website or through this SDK integrated into the app and then when you do those things and it observes that you do those things, it ingests that data into its own system and because it has an IDFA linked to your Facebook user account and the destination property had either the pixel sending data back with your Facebook ID linked, or the app had your IDFA attached to those events, it’s able to enrich its profile of you. Then it knows more about what you’d like to do and what you like to, more importantly, what you like to buy and then, with that knowledge, the next time it serves an ad to you, it’s a little bit smarter about what you’re most likely to click on and interact with.
I wrote a long thread on QuantMar, which is this stack exchange site that I operate. I found this video from, I don’t know, six, seven years ago from one of Facebook’s ad auction economists and he was giving a presentation at UC Berkeley to some grad students and it was on some grainy cam and I don’t know that it was meant to be seen by a lot of people, but he just walks through the whole logic of the auction algorithm, it’s really, really interesting.
The way they operate that calculus of deciding which ad to show you is an impression pops up in your feed and they have a whole universe of things that they could potentially fill that with. That pertains to content, so like stuff that your mom shared or whatever, but also some of those units are dedicated to ads, some proportion. The way that they decide what to fill into that space, onto that canvas, is they take this whole set of content options, whether it’s content or an ad and they look at your profile and your history of clicking on stuff and they say, “What’s the probability that this person’s going to click on this thing?” Then they multiply that times the expected value of you going to do that thing and so if it’s content it’s “Are they likely to engage with this? Are they likely to leave a comment? Are they likely to like it?” Engage with it in some way beyond just scrolling past it. If it’s an ad, “How likely are they to click on it? What’s the expected value of their participation in that product? Are they likely to make a purchase? Are they likely to put something in a cart? Are they likely to complete a level in a mobile game? What’s the expected value of that?” And then they use that expected value, they moderate it by the click probability and that’s how they rank them, that’s how the whole bid mechanism works.
So if I’m a niche product, and I have a very deep monetization in my product, or a very early-stage monetization where it happens very quickly but there’s a low probability of click, I have to bid a lot higher to win that space than somebody else that’s got lighter monetization but there’s a higher probability that you do the click and then do that thing subsequent to that click. So that’s the kind of logic that’s used to power the filling of those impressions, those content spaces. That’s the logic they use and then when you do that thing, well then that helps them inform the next decision.
And that’s the biggest input, right? What did you click on previously? That’s far more signal than any of your other behavioral data, what are you actually doing and clicking on —
ES: — and then, ideally, purchasing, right? That’s where I’m going.
Right, we’re going to regress to last click in some respects, whereas the power of being integrated into the website you’re going to, or into the app you’re going to, is this person didn’t just click on it, they went all the way through to conversion, or they went all the way through to buying gems in the game, or whatever it might be. That’s the part that’s getting lost.
ES: Exactly. And so to just shift back to the question about what does this all look like when you don’t have the ability to observe those conversion, when you don’t really the ability to calculate that expected value as discretely as directly. Also, you don’t have any ability at the user level to do that at all, so what does that look like?
We see where the directions these machines are lurching in. Snap talked to a lot on the call about Advanced Conversions, and so what they’re doing there, and they don’t have public schematics that you can investigate or GitHub files that you can pore over or code, but the way they describe is, well, they have all the click data because that all happens within the first party environment. They know exactly what you click on, what you interact with. You, the website owner, the Shopify merchant, the app developer, you have all your conversion data. Now if you scrub both of those datasets of identifiable features, you can join them in a way where you might be able to say, “Well, this click happened at this time and then, well, what do you know, an install happened 30 seconds later, that’s probably that same person.” If you’ve got anonymized enough data then it’s pretty private, but you can link those together. The downside of that approach for the advertisers, you’re giving all your data to these companies because you have no idea where the user came from, that’s the whole point of ATT.
That’s the thing that is so striking to me about this. I’ve written a lot about Facebook Shops, for example, where if you can move the e-commerce onto Facebook then they can track everything from the ad to conversion but I think this point about the data is super-important as well. Facebook, in particular, has really been trying to push advertisers to just trust us and you go in and their offerings are more and more advanced, which for someone that’s a very basic advertiser is great. You just go up to Facebook and say, “Look, this is how much this customer’s worth to me. This is how much I’ll pay for it,” and Facebook will do everything else. Of course, in exchange for that, they take much more margin of the ad as opposed to you doing all the work yourself, but especially the smaller you are the better a trade-off it is.
Now, if Facebook advertising’s going to work, that’s going to be the only option even for big players, you’re just going to have to trust Facebook that it works — that seems like the obvious takeaway to me. You’re just going to have to give it all to Facebook, Facebook’s going to make these massive investments in matching these datasets, and this huge CapEx increase on things like for data centers to basically apply machine learning to this matching problem. If you’re an advertiser on the outside, what other alternative do you have?
ES: Facebook published a blog post a few weeks ago where they go into detail about some of these new initiatives, these new measurement and attribution initiatives that are aligned with the privacy-centric approach that it was forced upon them with ATT, but in general it is the direction that everything is moving in. ATT’s not the only impetus for any of this, there’s the deprecation of third party cookies, this is happening across the board, it’s not just the Apple decision. We talked a lot last time about the motive for the Apple decision but, in general, the digital advertising environment is moving away this user-centric tracking.
I think it’s interesting because the cookies stuff has been happening for a while, but it’s been a slow boil, whereas ATT basically took all the changes in the cookie universe, compressed it into one quarter, and then also jumped ahead massively. It was the same sort of changes but it happened very rapidly.
ES: Because Apple owned the platform and could make that unilateral decision, whereas with cookies it’s more of a consortium of companies that have to agree on a solution. This is the direction that things have been moving in for a long time, ATT was a very accelerated shift, but nonetheless, this kind of thing was happening.
Facebook describes some of these efforts and one of them is called Private Compute, and I might be some of the details about this wrong, I have a superficial grasp on it, but there’s this server that multiple advertising platforms can contribute their data to, so the advertiser and the platform data gets mashed on the server that almost exists as a data co-op. So the matching happens outside of the data environment for any one of the platforms and it happens in such a way where there’s unbiased logic that’s applied to the attributions and then what happens is the attributions for the conversions get aggregated and then sent back to the platform, so they contribute all of their data in a triangular way to the server, the matching happens, and then they receive back the aggregated campaign performance data without knowledge of any other platform’s data. That seems like a really elegant approach, but I don’t know how feasible that is, that sounds like a massive undertaking.
This is very interesting because I noticed this, I can’t remember if I wrote about this, but where for a while there was a lot of tug and pull within the advertising ecosystem — there’s all this money to be made, and so you’re fighting over who gets what share of it and should you have a third party verifying it or all these sorts of pieces. Then ATT comes along and everyone’s like, “Okay, we all have to be friends now. We’re only going to get through this together.”
The big question I have and this in the long run, five to ten years out, the idea that ok, there has to be a consortium, these companies have to figure out how to work together makes a lot of sense, but when you start getting into the brutal efficiencies that are necessary when it comes to advertising, where you’re talking about fractions of a penny and measuring your ROI to as fine a detail as possible, doesn’t this play out in the long run that Facebook just becomes that much more a center of gravity? The consortium sounds good in theory when all these companies existed, but if you were to start from scratch none of these companies would exist because their reason to exist and their linchpin is gone. So it’s like newspapers on the Internet agreeing to collaborate, when actually in the long run the New York Times is just going to hire all the journalists and dominate. It feels like that’s going to be the outcome here, or this too bullish on Facebook’s behalf?
ES: I agree with you. It’s a very elegant, but also technologically challenging solution. It seems like a more straightforward path for both advertisers and platforms, what I call the content fortress, is what you just described, Facebook Shops. Like, “Hey, we could build this data co-op and everybody contributes their data and some neutral third party does all the matching and we can all trust them”, but these ad platforms aren’t really trustworthy in general.
(laughing) Much less trustworthy than Facebook!
ES: Or, instead of doing all that, why don’t you just bring your shop onto Facebook, just sell on Facebook and then we’ll just observe the data and we can target ads with that data. That seems like, “Hey you, Shopify merchant. Doesn’t that sound easier than going to all this trouble?” It just feels like that’s probably the more direct path to the future.
How many games do you think are going to switch to — because Facebook has experimented for ads — where instant play games where you can, in an ad, can start playing a game and then you can convert from within the HTML5 streaming experience to an install. Do you expect them to lean into just like no, you’re playing within Facebook?
ES: That could happen, so I think if you look at Facebook’s top advertiser categories, and Sheryl enumerated these on a call a couple of quarters ago, but it’s retail, it’s e-comm, and then I think third is probably gaming. So I think they’re starting with the biggest and the most important. She even acknowledged on this last call that, “Yeah, e-comm and gaming were hit the hardest” which means they were making the most use of these kind of IDFA-based profiles. So I think if you start at e-comm it makes sense, I mean, that was an issue that was already started during COVID anyway, just because leaning into the acceleration of retail spend on the smartphone. So you start there and you move these transactions onto Facebook, do the content fortress thing with e-comm.
Gaming’s more difficult before there’s a precedent for Apple not allowing that. The patient zero of this was an app called AppGratis, and if you’ve never heard of this it’s an interesting story. They had an arcade app and it was free, AppGratis was free, French company. They were doing really well and what they did was the way that you did games marketing in the early days of the App Store. This app was like an arcade, it was a collection of games you could download. You could buy the top promoted spot for that day in that app, and that was how you did user acquisition early on in the early days at the App Store. One day Apple just said, “Nope. That’s an apps store within an apps store,” they blocked the app update. They said, “This app can’t exist. It violates our terms,” because the App Store clearly says, the developer guidelines say, “No apps store within the Apps Store”. So that was the end of the company, basically, they raised quite a bit of money. They pivoted but that was the end of that incarnation of the company. So there’s precedent for them doing that and I think they’ve held that line since then.
My question is, they made this big unilateral platform policy change, and I’ve maintained that that was the only bullet in that gun. They could do that and after that they had utilized their opportunity to make a big sweeping policy change or enforce policy in a very draconian way. I wonder what happens if companies just stop complying with ATT or even stop complying with some of these App Store guidelines and they said, “You know what? I’m going to build an arcade app. What are you going to do? Are you just going to rip it out of the App Store? Then we have Epic versus Apple again” or some form of that.
I just wonder what is the cost of not complying? Can Apple enforce some of these policies in the same very severe way after ATT has brought a lot of attention on the way that they operate their platform. Maybe they can. I do feel like you’re going to start seeing some of that with the ad platforms at the very least, they’re not going to be compliant fully with ATT because SKAdNetwork just doesn’t work. This measurement solution that Apple has given in replacement of the IDFA doesn’t work.
You’ve just touched on two of the things I want to get to. So, you get to choose. What do you want to do first, talk about fingerprinting or talk about SKAdNetwork?
ES: Let’s talk SKAdNetwork.
Why is SKAdNetwork so bad? What are its limitations in, I know there’s a lot of them, so it’s hard to summarize it. But in a nutshell, what is the problem with SKAdNetwork?
Just for the audience, SKAdNetwork is Apple’s API for apps to basically measure to do ads. It’s supposed to be a replacement for what they got from Facebook, etc. It’s from Apple, Apple has perfect knowledge of an app because it’s happening on their OS and the app was obtained through their App Store and so Apple could theoretically make SKAdNetwork actually even better than anything Facebook could do because they have perfect knowledge of what’s happening, but it’s definitely not that. What are the issues with it?
ES: It’s hard to know even where to start. First of all, SKAdNetwork was actually released in 2018 and I discovered it somehow on a message board or something for app developers. I wrote an article about it, like basically, “What is this thing? Is Apple going to totally transform mobile advertising with this SKAdNetwork?” And nothing ever became of it. It was totally irrelevant, no one used it, no one —
It wasn’t a competitive product.
ES: I think it was just they were releasing V1 in anticipation of ATT. I don’t know that they necessarily had planned ATT in terms of the exact way that it was rolled out and exact way that it was defined, but I think they knew they were going to do something like that down the road and they wanted to have this ready. Anyway, so at WWDC20 they introduced the Version 2.0 of it, which was designed as the replacement for IDFA-based attributions. What it does is it allows you to attribute an install to a campaign, but without including any information about the specific user that installed the ad. It’ll basically provide you with the knowledge that an install happened and that install was driven by some campaign, but you don’t know which user that was. It can include some context around what that user did in the app subsequent to installing, so there’s ways to instrument events that can be tracked and returned with this SKAdNetwork signal, it’s getting a payload which is called a postback.
But it’s just designed in such a way to be totally inadequate, I think that by design it does not work. There’s a couple of reasons for that. One, is that operates on this timer system that is incomprehensible as to what the purpose of this timer system is, aside from just making the pushback data useless.
Just to make it clear what this is, it’s a random timer where — I guess the theoretical what Apple would say is, “We want to make it completely impossible for you to tie an install back to any particular user by going through your logs and figuring out exactly when or if something happened”. But the reality is, is one, no one’s going to do that because why would you, no one actually wants that information and it’s tremendously difficult. But then two, if you can’t actually know the timing of when something happened, you can’t actually understand if the ad was performing or not.
ES: Exactly. You can’t understand the economy of the campaigns because you don’t know when the installs happened and if you make changes to the campaigns, then you can’t track that.
Right. Was it attributable to the previous campaign or this campaign, what’s the connection there?
ES: There’s two timers actually, there’s one timer that’s minimum fixed 24 hours and then one timer that’s random up to 24 hours. So the kind of minimum turnaround is between 24 and 48 hours. The more context you try to capture with that postback, the longer the first timers are reset-able. So every time you try to capture another event that is considered higher value than the previous event you had captured than the timer resets. It’s just very convoluted, and so it basically makes it impossible to cohort this data, it makes it impossible even to know on what day an install happened.
So you can’t say, “I ran this ad and got this group of customers and I don’t know who any of those customers are” but I could say, at least in theory, “This cohort performed better than this cohort, which means this ad works better than this ad or this targeting criteria works better than this targeting criteria”, but actually, none of that even works.
ES: Right. And then that’s compounded by the fact that they limit the number of campaign IDs that you can utilize within any ad network for promoting any given app to 100. The problem that introduces is, there’s no creative parameter in the postback, it’s just the campaign ID. The way that a Facebook or a Snapchat or any sort of kind of modern ad platform works is you’d have a lot of different creatives attached to any campaign and those creatives have budget distributed to them based on performance. Well, you used to be able to measure the sort of performance of the individual creative in real time.
Now, it’s only at the campaign level and you only have a hundred campaign IDs so that means, for any of these platforms to do this, first of all, you have to keep a bunch of campaign IDs in reserve because anytime you make a change to a campaign, you’ve got to shut that campaign ID down for between 24 and 48 hours because postbacks are still going to be coming in because of the timer system. But then all the different combinations of targeting and ad creative, they all require their own campaign ID, every single unique combination of those things requires some campaign ID.
So for example if you’re running a campaign on Facebook for a game install, you could give Facebook ten different creatives and Facebook will not just run all of them, they will dynamically in real time, increase the use of the one that’s performing the best. This is all automated, you don’t even actually, Facebook will take care of all that for you. Apple on the other hand, you have to create ten campaigns using ten of your IDs, you have to run all these with these variable random timers. You have to wait at least X amount of hours to even know which one’s performing well, then you have to wait an additional 24 or 36 hours for that campaign to flush through so you can focus on the one, and this is a massive increase in work to do because it’s not automated at all but also, because of that one hundred cap, you actually, you should run fewer creatives and just sort of roll the dice more.
It’s like we got to this world of “Advertising works, half of it works, half of it doesn’t. I just don’t know which half.” to “Actually, we can know what works and increase the spend toward it” and the thing about this is, none of this is compromising user data at all. It’s all about the creative, literally what the ad looks like. Why is Apple killing that too? Grant them all the privacy arguments, what is the privacy argument for doing this?
ES: I don’t know. I mean, I don’t know, because their own reporting API does include a parameter for creative —
That’s a good question, so Apple has their own API, if you go through Apple, the App Store ads, you can run variable creative, right?
ES: Right. So Apple’s own ad network, which is Apple Search Ads, it doesn’t use SKAdNetwork for reporting, it uses its own reporting API called Apple Ads Attribution API. First of all, the reporting postbacks are not subject to a timer system, and there is creative content in the payload, so they obviously recognize that these are things that advertisers need or at least want, and they provide them to them with their own ad network but not with SKAdNetwork.
Let me go back to SKAdNetwork real quick, because there’s kind of one other big problem with it. First of all, they pass all the postbacks through a VPN, so you can’t geolocate the postback, you don’t know what country it’s coming from, there’s no country parameter.
Which, the initial response is, “Oh, that sounds great, that sounds very privacy preserving.” But what’s the problem with that?
ES: Well, that just creates the opportunity for fraud to happen. The ad platforms can target, you say, “Hey, target people in the US” and they start targeting people wherever and the postback doesn’t contain any of that information, you just see the installs coming from that campaign so they must be from the US, right?
Could be a click farm in Vietnam or something like that.
ES: Right, exactly. Then there’s also just a privacy threshold that it applies to that in-app context that I was talking about, that you can capture some in-app context around different events that the user did and it is called a conversion value, but they applied this privacy threshold to that, which does actually make sense. I don’t have a philosophical problem with this, it’s just the implementation. The privacy threshold says you have to have generated enough of these events before we start actually including them in the postback. Because let’s say that you had one event, let’s say that it was, I’m operating a game and it was the buy the $10,000 IAP, just as an extreme example. Well, there’s probably only going to be one person that does that every so often, it’s a pretty rare thing to do. And so if I got the postback for that very rare event, and there’s only one person that does it every couple days, and I know who that was.
Yup.
ES: So I get that, that is privacy preserving. And I understand why they do that, but they haven’t really made the logic of that public and the value apparently changes over time, because there was just this measurement company that most advertisers work with, they just reported, “Oh yeah, we’re seeing a much higher prevalence of null values there, something changed, we have no idea what it was”. It was just the setting was changed and no one really knows how that operates, it’s just impossible to use. The combination of all these things make this impossible to use for measurement and that’s what Snap was talking about, it’s just the total deficiency of SKAdNetwork was the source of all their frictions from ATT.
So here’s the big question. I think that there will be Apple defenders that say, “Look, it’s new (even though it’s not new, 2018), but SKAdNetwork, they’re going to get better, the Apple API works better because Apple’s dogfooding it first sort of internally.” Do you see that happening? Is SKAdNetwork getting better? Or is it actually getting worse?
Honestly, it seemed if Apple would just give outside developers and outside advertisers the same ad network API that they use internally, a lot of these problems would not be solved per se, because nothing’s going to be as good as the IDFA, but you could actually start doing intelligent targeting based on creative for example, you could start getting more feedback. Do you think Apple will move in that direction? Or is there going to be two paths here which is Apple’s solution versus the other solution and that one’s just going to stay pretty crappy.
ES: I’m not very optimistic that SKAdNetwork is going to improve. Now, they have rolled out another version where they included for instance, view-through attribution, which that was objectively an improvement. But Apple just historically, has not been very good at improving their developer App Store analytics tools over time. Like there’s iTunes analytics, it just hasn’t really improved much since it was released how many years ago. I don’t know if it’s designed, I don’t know if they want it to be good, I don’t know if they want it to be useful, I don’t know if they want it to be something that is sort of functionally competent as a replacement for IDFA while also being privacy protective.
Look, I’m glad the IDFA was expunged, I think it was abused, I think the way it was used was exploitative and users had no control over it, they had very little control, they didn’t know they had control over it. I like the idea that people have choice, but the way that they’ve implemented SKAdNetwork is just not useful. Tim Cook called the App Store an economic miracle, why is it an economic miracle? Because of direct response advertising, because you gave people control over the growth of their products and you just handicapped them.
So where does this go in the long run? I wrote an article last summer making this point, that Facebook and Apple are super symbiotic, Facebook did a lot of the dirty work for Apple as far as the App Store becoming the monster that it became.
So where is this going forward? I mean, does Apple’s ad network and their superior API just completely take over gaming in general, they increase inventory? Does Facebook’s ability to, again, where it’ll take them longer to recover because they were so good, but when they recover, their position is even more impenetrable, where does the long run play out for this? Is Snap going to be able to come back? They got hurt less because they’re less built up, but they’re still much smaller than a Facebook, for example. Just stepping back, where is this ecosystem going?
The other piece to add in too is, this has only happened on iOS so far, everyone’s waiting for the other shoe to drop on Android and right now a lot of money flew to Android because at least we know how it works over there, but when that happens…
ES: So I’ll start with that. I don’t see the other shoe dropping to the floor, I don’t know, to lean into the metaphor. Android’s doing a lot of stuff around privacy, but it’s more about just informing users. They’ve got the privacy notifications in Google Play and they created the private compute environment on Android, like the sequestered memory space to do on-device computing, the stuff that we talked about in the Twitter Space a couple months ago. They have done stuff that’s very privacy forward, but I don’t think they eradicate the GAID at least not until they have a FLoC type solution to put in place.
So what they’ve announced is they had a setting in Android that would basically limit ads personalization, but it didn’t really do anything. So when you turned it on, basically it just limited how your behavioral data could be used to target ads to you in Google Play, so that was Google’s own ad network placement and so that limited that, but it didn’t limit access to the GAID and so what they said was that, we’re going to rewire that, so when you turn that on now, it’ll zero out your GAID. So it’ll basically be functional equivalent to the setting limit ad tracking that Apple had, that was a predecessor to ATT, as a system level-
Which is around for several years, I think, something like a third of users had opted into it.
ES: Like 40% in the US. So, I mean, there was opt-in, there was uptake. Then that’s the trajectory there. It was introduced in 2016 and by 2020, it was 40%. So over the course of four years, I don’t know, it’s 10% a year. So I’d imagine that you expect that trajectory to be similar in Android, after four years, 40% of Android devices have no GAID. I wouldn’t be surprised if that was the extent of it, I don’t think that they’re going to be as aggressive as doing something like an ATT.
Does it feel like because ATT has gotten a lot of negative pushback, that maybe Google feels like they actually, originally, they’re like, “We’re going to have to follow Apple on this” but now maybe they’re like, “Well, we can probably get away with talking around it and actually keep it in place”
ES: Well, negative pushback from people in our Twitter circles, but I don’t know if it’s been universally negative.
Yeah. That’s true.
ES: They got a lot of praise in the sort of mainstream press.
(laughing)Us two are like a bubble of two
ES: Well, I mean, even within my Twitter circle, I don’t think everyone agrees with me. In fact, I know that to be true. I think Google is also just an ad network, at it’s very heart as an ad network. I think Google, Apple Search Ads, I mean, there was the post on the Financial Times, they’ve clearly gained some market share here, but they’re not an ads company, they’re never going to be an ads company. That, if anything is just sort of some icing on the cake from ATT, I don’t think it was the primary motivation. I don’t think they’re going to make that much money in advertising even in five years. There was the Evercore analyst who projected $20 billion in revenue — for Apple, what is that? That’s a rounding error.
This is interesting because, this is what’s kind of weird about this, because the fact that Apple has an ad network makes this so brazenly anti-competitive. They’re destroying other ad networks, and they’re giving an alternative that is markedly inferior to what they’re using for themselves. We’ll link to this article you wrote about ATT, but you have this chart in here about inequitable advantages to Apple’s ad network created by ATT with proposed remedies, and you read it, it’s jarring. It’s like, how can they do this?
But to your point, I think your theory sort of before, is that Apple just wanted more control over apps, they resented and didn’t like the fact that Facebook actually controlled what apps got installed, and this was much more about control and that almost makes more sense because if this was really about being anti-competitive, they would actually have more shame, it feels like, or at least more legal wariness about it, but it’s so brazen. It almost feels like it can’t be anti-competitive because no one would be this stupid.
ES: I one hundred percent agree. I don’t think it’s about the ad revenue, that’s a sort of minor victory. I think the sort of biggest motivation here was just to control discovery, to control which apps got installed on devices and how. And if they do that, they sort of maintain that linkage.
My sort of bigger, I don’t know, maybe too abstract of a theory, is that they’re pushing back on the cloud based future, maybe they’re pushing back on the metaverse, because in the metaverse, it doesn’t matter how you access it. The iPhone, Android, Samsung, TV, Facebook Portal, all these things are entry ways, entry points into the Metaverse, and I think that’s what Mark Zuckerberg said in your interview.
If you look, there’s that email that he sent to his top execs in 2016 or something, they were kicking the tires on Unity and they said, “Look, we’ve got to break this platform dependency. We have to own the next big computing platform. If we don’t, we’re going to be in this position again where we’re totally dependent on them and we’re totally at their mercy.” This reinforces that dependency. The constriction of the way that you discover content, because if you reinforce that one-to- one relationship between “I have an iPhone therefore I use the App Store to discover content”, then you delay that metaverse future for however long because you need a way to enter the metaverse.
Now, the other piece of this is, well, how does Private Relay factor into that? Because, at some point, Private Relay breaks the connection of your phone to the broader open Internet and does Apple at some point get to become an actual gatekeeper?
Like hardware, the ISP level. Apple’s like your ISP.
ES: Right, because they’ve told Google Stadia, Microsoft’s Xbox Game streaming service, “You’re not allowed on the App Store. We’d have to vet every single one of the games in your streaming service, and unless we can do that, you’re not allowed. And so therefore your service won’t be available in the App Store.” But what happens if they say, “You know what, we’re going to block access to that via Private Relay?” Maybe that’s extreme. I mean, that would be very, very heavy handed, but they could do that, that gives them that kind of control.
Just on this Apple point. One is they just want more control and we’ll grant them the privacy argument. And to your point, the IDFA thing where you can measure down to the individual, that’s the point of it, great for small businesses because they don’t have a lot of conversions to build large cohorts. I think it’s one of those things where Facebook’s just the worst messenger even when they’re right. This is really bad for small businesses because you just don’t have the data set or the number of conversions to build an effective cohort strategy, the fact you could do it on an individual level is much better. But we’ll leave that aside for now.
You talked a bit about fingerprinting and this idea of using IP address and model number and OS, and you can get a good idea of who someone is. It’s not near as good as IDFA because it’s not as persistent, it falls apart pretty quickly. But that Apple’s not really enforcing that and the only way they could enforce it is via the App Store Review Team where they look at an app and see, is it pulling these parameters, etc.
This leads to two questions. Number one, do you think Apple will crack down on fingerprinting? Number two, if they don’t, does it make this whole ATT thing feel like it’s just about Facebook? It’s really just about Apple versus Facebook?
ES: Yeah. Well, I mean, I think even if they do crack down on fingerprinting, my thesis remains the same, that it was all about Facebook. It just makes the sanctimony of this just overwhelmingly phony. So with fingerprinting or call it probabilistic attribution, which is what the people that do it call it.
“This is probably the first that installed it.” Yeah, love it.
ES: This is probably the first that installed it, because it’s the same IP address, same OS version, same phone model. So you’re able to attribute the install, but not anything else because IP address changes and just —
Especially on mobile, if you’re moving around.
ES: Right, exactly, and the validity of the fingerprint degrades pretty quickly, and so maybe Apple just doesn’t care about that. I think they do because they explicitly stated in ATT guidelines, “Fingerprinting’s not allowed. You may not collect device perimeters to uniquely a identify device.” Now, the companies that are doing this say, “Well, we don’t uniquely identify it. We just have a probabilistic estimate. We have some probability that informs our belief that this is the user that we saw click and ad just 10 seconds ago.” And so, fine, whatever. I mean, that’s semantics. But the fact that they’re allowing that to happen when they explicitly stated that this violates ATT guidelines, it undermines the privacy motivations, the purported privacy motivations, here.
Now the question is, will they police it? I don’t know. I thought they would’ve in iOS 15 because that’s what Private Relay does, it prevents fingerprinting from being possible in Safari. So you’d think that they’d want Safari and the treatment of Safari, the treatment of mobile web traffic in their browser and the treatment of app traffic to be the same, but they didn’t apply it there. Now, Private Relay is probably pretty expensive to operate on the scale of all apps and so maybe that’s the reason. But still, you’ve got this mechanism that does what you say is necessary in order to protect user privacy and you’re not applying it to apps.
Right, because HTTPS traffic from apps does not go through Private Relay. It still goes directly to the server.
ES: The vast majority of all traffic through apps, right?
Yep.
ES: So anyway, probabilistic attribution’s rampant, but it’s not something that’s really possible to do if you operate the app in which your ad network traffics impressions. Because, well then, Apple can punish you directly. They say, “Hey, look, your app is polling this data, this data is probably being used to fingerprint. We’re going to stop allowing you to update your app until you stop pulling that data.” And then you have to update your app and stop doing it. But if you’re using an SDK that does it, like a third party SDK, well then, in order to police that you’ve got to punish the app developer and you basically have to punish every app developer because every app developer has one of these SDKs integrated.
Right. So what’s an example of an SDK? I was going to do MoPub but I guess Twitter sold it. We’ll use MoPub. So if you have MoPub in your app and it’s doing fingerprinting, Apple would have to block your app that has nothing to do with AppLovin, who MoPub got sold to. Will Apple really bring the hammer down on literally hundreds of thousands of apps that have these SDKs installed, which is really the only way to handle this?
I think this gets to the real limitations of Apple’s approach because they can take away the IDFA, that’s a technical solution, but so much of ATT relies on App Store enforcement, and it just doesn’t scale. It doesn’t scale from a regulatory perspective because the more they do, the more it brings attention to them. But also, it doesn’t scale from a “Are we going to punish every app developer for the sins of an SDK” perspective either.
ES: Right, exactly. So, first of all, there’s a precedent for doing just that, I mean, they did that. A couple months before ATT there was an ad tech vendor that I guess they determined was doing fingerprinting or they were collecting too many device parameters and they said, “Hey, every app that has that SDK integrated, your app updates are going to be rejected until that SDK fixes this problem,” until they reform this violation. And they did that and they blocked app updates and the SDK issued an update within a day or something, all the app developers had to reintegrate or had to integrate the new version and then the app updates got approved. So there’s a precedent for doing that, but not at the scale of every single app.
Every single app that operates a business has one of these SDKs in it that would violate this fingerprinting clause. There was a Washington Post article a couple months ago where some security firm just did an analysis and they looked at some number of apps, and every single one had an SDK that did this fingerprinting. And they ran the gamut from ad tech measurement partners to just ad networks. Everything, everything was doing it.
To be clear, every app is doing it, including I’m sure the Washington Post app, which is always fun when you get lectured about privacy by these companies that have the most trackers of anyone.
We could talk literally for hours about this. I think I find this absolutely fascinating, both how advertising works in general but all these big changes that are happening. Are there any final thoughts from your perspective? I think we should definitely check in again in a few months particularly once the next couple of earnings calls I think for Facebook and Snap in particularly will be very interesting. But is there any takeaway other than at the end of the day it almost feels there’s going to be this little scurry in the undergrowth where these ad networks are going to get away with stuff and Apple’s not going to police it all. The pain’s going to be felt the most by Facebook, but in the long run, any company that has just a ton of data and a ton of customers, the gravity of that data is actually only going to increase. And so in the long run, the Amazons, the Googles, Facebooks — Facebook in particular short-term hurt, Amazon’s probably helped a lot — but long-term, more strategically secure? Is that the takeaway you have?
ES: I think so, I think this is a very exciting time. You’re seeing the boundaries on a lot of these companies that I think were considered inviolable at some point just being shattered and there’s a massive opening up of opportunity to grab market share across various sectors.
Is there a chance that Facebook doesn’t actually come back from this, this ends up being more fatal than we think?
ES: No, I think where I see Facebook moving is I wrote this Twitter thread about it last quarter and my sense is, short term they’ll all regain measurement efficiency, there are approaches. Instead of the brute force guerilla collect all the data and attach it to a user approach, just much more sophisticated and technically demanding statistical and probabilistic approaches. And they’ll move in that direction.
That’s all the CapEx on computers to crunch all that data.
ES: Right. Exactly. In mid-term, it’s content fortresses. They just pull as much content interaction into their own environment as possible. It’s all first party data to them and they just get stronger as a result. And then, longterm, it’s metaverse, that would be the bull argument for Facebook. But they have risks, this is unprecedented stuff, this is corporate warfare, like atomic warfare, this isn’t skirmishes at the border.
Right.
ES: But there’s a whole lot of new opportunity that’s arising as a result of this for creating whole new measurement paradigms, for potentially shifting a lot more content onto the open web. So maybe this just reshapes our relationship with content and reshapes the things that we interact with on our phones and elsewhere. It’s also a very exciting time because just these ideas that we held as absolute have been shattered.
And it’s not just on Facebook, that’s on Apple too. The idea that you could link out to an alternative payment system in your app a year ago, two years ago, you’d be risking a lot to do that, that was almost unthinkable. And now, it’s policy. The idea that you could just with aplomb email your users and say, “Don’t buy your stuff in the app, buy the stuff on the web shop right now.” that’s totally new, these are new commercial venues that have opened up. It creates a lot of opportunity.
It was very interesting actually on Twilio’s earnings call, they were talking about this. The ability to connect directly to your customers, it’s always been valuable, but now it’s astronomically valuable and that’s why they feel well placed for obvious reasons. But I thought that fits with what you’re saying here.
ES: Yeah. At the same time that Apple’s exerting more dominance in some areas of mobile, it’s also having its dominance receding. One thing that’s really exciting about the opportunity to open up a web shop that services the in-app content environment, is that the App Store really sucks as a storefront. It’s just clunky. There’s not a lot of options for it, there’s not a lot of tools, you can’t A/B test stuff. If you have a web shop, you could do anything. The web is super easy to just quickly make changes and deploy them, you don’t have to do an app update that requires a manual review. You can do everything on the backend, you can A/B test to your heart’s desire. It’s just much more dynamic than what you can do in this very restrained environment, which is the App Store.
So I think the ability to have a web shop where you personalize offers to people, you offer dynamic discounts, you offer a totally tailored bespoke product to you based on your behavioral profile that I have from a first party data environment, that’s super exciting and that didn’t exist in the App Store before. That’s an opportunity that just got opened up and really, I mean, to do it in public without having to worry about Apple punishing you, that opportunity only arose in the last couple weeks.
That’s super interesting. So in your estimation, if Apple has to allow purchases outside of the app, then net relative to killing IDFA, Apple’s feeling the pain actually probably much more?
ES: Well, we’ll see. I mean, I don’t know. Maybe that’s what IDFA was all about. Maybe it was just, let’s try to wrap our arms around this and say, “Hey, you could do a web shop, but you need featuring from us to be a successful business. So do you really want to risk that?” Maybe that was the whole idea here. Maybe it was just pre-empting what they saw as the inevitable opening up of the App Store environment.
Well, it’ll be very interesting. This has gone way longer than a typical Stratechery interview, but there’s so much about this area, I agree with you, it’s absolutely fascinating. We will definitely have to touch base again and see how stuff continues to roll out. But thanks for jumping on, Eric. It was good talking to you.
ES: Thanks for having me.
This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery.
The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.
Thanks for being a supporter, and have a great day!