Good morning,

I am excited to welcome Eric Seufert back for a Stratechery Interview about the evolving digital advertising landscape in a post-ATT world.

Seufert is the creator of Mobile Dev Memo, where he writes about mobile advertising; Seufert also invests in startups in the mobile space through his fund Heracles Capital. Seufert previously worked for several mobile gaming startups, including as the Vice President of User Acquisition for Rovio, and was a consultant for multiple companies exploring the mobile space.

In case you misseed my previous discussions with Seufert, we talked in February 2021 a few months before Apple’s App Tracking Transparency (ATT) policy came into force; our second was last fall just as the impact just started to be felt; our third was in May after Q1 earnings showed the first major signs of impact. Now that Q3 earnings have passed I wanted to check in with Seufert not only about where the industry is today, but also spend some time looking forward to where things are going. This includes covering Apple’s big overhaul of SKAdNetwork, speculating about a possible Apple Ad Network, and why we’re both optimistic about Meta’s outlook. In addition, we discuss Snap’s disastrous year, and the Unity-ironSource-AppLovin acquisition drama.

As always you can listen to Stratechery interviews on your member-only podcast feed; to get your feed for the client of your choice, including Apple Podcasts and Spotify, please visit your account page.

On to the interview:

An Interview With Eric Seufert About the Future of Digital Advertising

This interview is lightly edited for clarity.

The Snap Disaster

Eric Seufert, I might have said this the last time, but you are definitely the number one Stratechery guest at this point and it makes sense given that the biggest ongoing story in tech continues to be the upheaval in the ad ecosystem. I know we’ve spent multiple interviews talking about ATT in the run up and then in the middle of it, but we’re still wrapping our arms around it and I wanted to get your updated view of the impact after the latest round of quarterly earnings.

Eric Seufert: First of all, thank you for continuing to invite me back, I really appreciate the opportunity to chat about this with you. My take on earnings — we always get Snap to tee things up for us.

(laughing) I know! I want to talk about Snap in a little bit, but you can go first.

ES: This quarter was just a disaster for them. For context last quarter in the Q1 earnings they had provided reasonably optimistic guidance for Q2, a month after that print, they had to revise guidance. [CEO] Evan [Spiegel] went to the JP Morgan conference and he said, “We’re going to miss guidance. We had to revise guidance down.” They filed a letter with the SEC, they sent an internal email around blaming macro factors. This was right before WWDC and they did in fact miss the guidance that they had provided in Q1. They missed on the top end, the bottom line, and they didn’t provide Q3 guidance. They said that Q3 looks to be about flat year over year, no growth I’m assuming.

They made a couple of points that I thought were really interesting and I thought are probably be specific to Snap’s business, but reveal a couple things about the entire tumultuous episode to my mind. One thing that Snap did early on, and this predates even ATT being introduced, but they downplayed the impact. They said, “Oh, we welcome these changes. We think this is a great development for the advertising ecosystem and it might present some challenges but we’ll overcome them.” Then in Q3 of 2021 had the miss. I remember I was on a plane at the time, frantically refreshing Twitter with a 30 second lag, they missed, the stock dropped 25%. They said, “Okay, we didn’t appreciate the magnitude of this. We didn’t appreciate how damaging this would be.”

I think we talked about it at the time. We were just like, “What world are they living in where they think this isn’t going to have any sort of impact?” Their whole journey’s been very bizarre because they had that Q3 miss then Q4 comes back, I have my theory about why this happened. And then, yeah, this year has just been a total disaster.

ES: Right, it’s just been very volatile, especially for them. But I think their messaging changes every quarter because in Q1, they said, “No, we figured out measurement, we figured it out. We’ve got advanced conversions, we figured it out. This is no longer really a problem, we’re going to claw back performance over time.” That put investors at ease because they had figured it out and then it turns out they did not and we saw what happened with Q2 and presumably Q3 is not going to be very good. First of all, Evan didn’t participate in the call, I guess he was there on the call, but he didn’t.

Yeah. Very weird.

ES: That was strange. But they said, and I’ve got this quoted, “Deceleration began with the platform policy changes implemented in Q3 of last year.” They called that out specifically this quarter. “Those policy changes upended the decade of advertising industry standards.” — very explicit there. Then they made a comment about third party tools and they’re talking about measurement and they’re talking about modeling conversions, I think we talked about this last time. That’s where Facebook’s made a lot of progress, they’ve managed to claw back some of the ability to not observe conversions, but to model them, to estimate them. In the call Snap was talking about this and they specifically called out the third party tools that advertisers used basically to keep an accounting of these conversions. Basically the analytics tools and the measurement tools — they’re talking about the attribution tools here and they said, “And those third party tools across both web and app have been impacted by platform policy changes. This has been particularly acute for web-based advertisers and has led to an increase in a relative emphasis on the same session, last click-based advertising where we need to show that we are performing, we’re still early in this journey.” They called out these third party tools, the tools that companies use for measurement. They integrate with all the ad platforms and it all gets aggregated in this third party tool.

Now that approach is what Snap had been alleged to have been doing by the Financial Times with some of these third party measurement tools in terms of linking data in a way that’s probably not totally compliant with ATT. These are just allegations, no one knows whether this is true or not outside of these companies.

Well, but I think you said something along the lines of Snap’s what’s it called? Conversions API?

ES: Advanced Conversions.

Advanced Conversions — I think you said this either to me offline or in an interview, that you were concerned about it, that the way they were presenting it as a solution, but it seemed like it was actually subject to the exact same limitations and it sounds like that’s exactly what has manifested here.

ES: Right, so they update guidance before WWDC and at WWDC Apple makes a very sternly worded warning about the use of these kinds of methods for doing advertising measurement, where the data is being passed around. Now, no single party has the entirety of the data, it gets linked up in this trusted third party neutral service, but nonetheless, the data’s moving around, and so I think I took a couple of cues out of this earnings with Snap leading the way. One is that, no, I don’t think any of these companies have the measurement stuff figured out. It’s that they were being foolhardy when they said that they did.

Second, I think there is a belief that Apple is providing new tools for advertisers and advertising platforms, we’ll talk about that later I’m sure — SKAdNetwork 4.0 — but they’re also going to be taking some things away. That there was probably this prolonged grace period where they did allow for things to happen that they’re not going to allow for anymore. Maybe that was the mediated conversion matching tools, maybe this was fingerprinting, but I think there’s a reason to believe that Apple’s going to give something, they’re going to take something away in this next iteration of ATT. What I also took out of this was that, in combination with macro factors, COVID overhang, and I’ve written a series about this, but we really are just in a digital marketing winter, this is just a very bad environment for digital marketers for ad platforms.

The Snap part is very interesting in particular. My theory on Snap is, remember that they foolishly IPO’d too early. They hadn’t yet built out their auction platform, all of their sales were their marketing team selling it directly to brand advertisers and then they go through this big shift and they talked about on earnings calls, “Oh, we’re moving all our folks onto our bidding platform.” That stuck with me because I think the easiest way to understand and explain Snap’s earnings, particularly the fourth quarter, where they were surprisingly up and everyone was like, “Oh wow. They really did figure it out” — my take is that Snap has said, “We’re 50/50 direct response versus brand.” I think a lot of what they call direct response is actually brand advertising that happens to be using their bidding platform. It’s not folks that are actually tracking conversions and so I think Snap in particular has been very hard to understand because they do have an intense amount of seasonality baked into the results much moreso than a Facebook, and so that’s why their fourth quarter is way up. That’s why the macroeconomic environment, I suspect, hit them more and made this last quarter particularly bad because that whole pullback is going to impact brand more.

But to the extent I’m right then my takeaway honestly is that Snap doesn’t really understand how their business works. That’s honestly the best explanation for why they’re all over the place. But to the extent that’s true, it’s interesting you mentioned this third party measurement point, because in a future ATT world, third party measurement tools really can’t exist. How are they going to get the data? How are they going to be able to verify? And so there’s going to be an aspect where you have to trust the platform to a greater degree, and I wonder who’s going to trust Snap? They don’t seem to even know how their own business works, and I’m going to rely on what they tell me about conversions? Or I can just go to Facebook, which whatever you say about Facebook, at least their stuff generally works, and if I’m going to trust someone, I’m going to trust them. Does that resonate with you? Does that make sense?

ES: Yeah, it makes a lot of sense. I think one thing, and it was just a throwaway comment. I don’t want to overthink it, but one of the Snap execs, I don’t remember who, said that, and this is quoting, “A significant majority of our revenue is direct response-related.” I think that is exactly what you’re talking about, it’s brand dressed up like direct response. This is not really direct response, there’s not a measurement intuition about it.

They’re not signing a contract like an ad executive, they’re using the bidding tools. But they’re using the bidding tools to do brand advertising.

ES: Exactly, and I think that set off an alarm, because I’ve read your stuff about this, and I agree with it. I think though, the other point here is that it’s not necessarily a post-ATT environment that we’re talking about anymore. We’re talking about the new legislative reality, the new rules for digital advertising will prevent third party measurement apparatus from just sitting in between an advertiser and publisher and consumers and just aggregating all that data. ATT was I think the tip of the spear, but there’s a torrent. Some of its already been passed like the DSA in Europe, but we’ve got ADPPA [American Data Privacy and Protection Act], which we could talk about later. The rules of the game are changing, the foundation of this entire economy is changing.

Well, basically just by and large, and I think this is an important point, we’ve said it again and again, but honestly it comes up in conversations, I think we need to say it one thousand more times — ATT is not just IDFA. IDFA is that device-specific number that you no longer have access to but it’s also all this legislative stuff. The big question is how do you see Apple policing that going forward? Are they going to kick Snap out of the App Store if they don’t abide by this? Because a lot of this stuff you could do server-to-server — the IDFA, Apple can technically limit access to that number. How do you see them, to the extent they actually follow through on these warnings, actually policing server-to-server data sharing and communications?

ES: So there’s a couple things. One is fingerprinting which, by and large, I think, becomes irrelevant with SKAdNetwork 4.0.

SKAdNetwork 4.0

Let’s start there. Walk me through SKAdNetwork 4.0 and the changes that are coming.

ES: SKAdNetwork 4.0 is basically like a page one rewrite of SKAdNetwork. It seems like they just burned down the schematic and built something totally new with a blank page. It’s better across the board and so all credit due to Apple, this is a welcome change, I think there’s a lot of potential here. I wrote a post about it, there’s a lot of detail.

There’s three big buckets of change — really four but three that I care about. Basically at a high level, big concept, SKAdNetwork 3.0 very much operates on this binary conception of privacy, you either surpass some threshold that Apple has determined for privacy or you don’t, and if you don’t, then you get almost no in-app context back in the postback that gets sent, you basically just get an install log but it’s sent on this random timer system and so you don’t get it in real time, you get it on this delay. There’s this privacy threshold which is not public, we don’t know if it’s dynamic, there’s a lot of mystery around it. But either you surpass that or you don’t in terms of the number of events that have been captured from a specific campaign from a specific app, and if you don’t then you get no context.

What they’ve done is they’ve added gradations now. It’s not just binary yes or no, but there’s gradations, and if you reach the highest level of what they’ve called now in the new terminology they’re using here — it’s not new but it’s new to this — is Crowd Anonymity. Basically the more people that you have in a group the harder it is to identify any single one of them, that’s the crux of that idea. So what they’ve done is, as you increase the level of Crowd Anonymity with the number of events that are generated, then you unlock more value from the system. At the highest level of Crowd Anonymity, you get very many source identifiers that pertain to the campaign and you could codify those however you want. You could get potentially creative information embedded in that and various parameters of the campaign information encoded in that. You can set it up however you want, but if you don’t, then you get just the normal 100. They’ve done the same with Conversion Values. Now, if you reach the highest level of Crowd Anonymity, you get the 6-bit options of Conversion Values. But if you don’t, if you just reach that middle tier and there’s three tiers, then you get what they call a coarse grained Conversion Value, which it’s just three values. It’s high, medium, low. But you could encode information in that it helps you understand whether that’s a good campaign or not.

Just to jump in on this because I think this is actually a really important point. Big picture, we were pretty grumpy at Apple last time, we could try to put on our “Let’s try to give the best possible explanation for what Apple is doing” hats here. I think the concern is if you are sending super specific feedback and you only have ten users just to use an extreme example, you can back out who specific users are and so what they’re saying is if you have millions of users we can send you very specific events because it’s basically impossible to figure out whoever every single one was. I just want to put a stake in the ground on that, because I think that it’s actually super critical to understanding broader changes in the ecosystem. We’ll talk about the Unity stuff later, but by implication you have to be large because if you’re going to get the best feedback, you have to have a huge number of impressions and events to make this all work. I just wanted to double down on that specific point.

ES: What I would add to that, I’m not going to dig into the details here because I think it goes beyond the scope of the conversation, but yes you just described it really well. But the problem was before, there was a chicken-and-egg problem. I have a campaign, and I’m spending some minimum amount of money against it because I want to measure the performance before I start increasing the amount of money that I spend against it. But I’m not getting any data from the postbacks, because I’m not spending enough money to generate enough events.

You don’t want to dump millions of dollars into that campaign, you don’t know if it works.

ES: I’m at an impasse, I don’t want to spend more money on it because I’m not getting any performance feedback, but I’m not getting any performance feedback cause I’m not spending any money on it — so, chicken-and-egg. Now with this, if I spend a little bit more money and I get to that middle level of crowd anonymity, I get the high, medium, low feedback, and I can encode that to mean something that is relevant to me to give me some signal as to whether this campaign could be a good one. Just having that intermediary step gives me a lot of really valuable context that prevents this from being binary yes or no and breaks me out of that chicken-and-egg problem. Now I could spend a little bit more money because I’m getting some signal that, “Hey, this is all coming back with the high value” and if I’ve set that to mean the predicted value of this user after ninety days, “Okay that’s high”, that means these are good users, I should spend more money, and it gives you the signal that you need in that early stage to determine whether you can spend more money against the campaign or not to really unlock all the value.

Are they also speeding up the feedback loop so that once you start spending high, you can immediately see if this is worth it or not before you’ve spent too much and make changes?

ES: Yes. The other thing that I’m really excited about is they just did away with the timer system, which was overly engineered, complex, and convoluted. Now there’s just attribution windows, and there’s three. So now you can potentially get up to three postbacks, it’s not just a single postback and the attribution windows are fixed. Now they’re not zero day, it’s zero to two days is the first window, but it’s fixed and so you know when you’ll get that postback. On top of that, because you know what the fixed interval is, you can actually model much more easily on top of that to get to install estimates whereas before you couldn’t, because the timer system was random.

The Privacy/Utility Trade-off

There is a saying that no lock on a door will stop a truly determined thief, but its simple presence will stop the vast majority. It feels like what Apple did this with the original SKAdNetwork was say “We’re going to try to stop the ultimate criminal who is going to apply gobs and gobs of computing to back out the most finely tuned information about a user”, and I think this new version of SKAdNetwork is where Apple is like, “We just need to stop the vast majority of bad actors who can do a trivial back out”. It feels like there’s been a bit of a wake up call, it’s like, “Oh yeah, we cut off the legs of this ecosystem. Our solution is Fort Knox and is completely untenable”. Yeah, it would’ve been better if they had done this from the beginning, but again, to their credit, as you noted, at least they’re doing it now.

ES: Two points there. One is you just described the core tension of digital privacy. There’s a spectrum and there’s a privacy/utility trade off and if you try to build something that’s totally impenetrable, you’re going to have no utility. There’s always going to be someone that has the motivation to try to get around whatever privacy protections you’ve erected. There’s always going to be someone. The question is what do you want to actually protect against?

There’s a approach to dealing with this called privacy budgets. So if there’s enough instances of me doing something, it’s easy to know it’s me. Not me “Eric”, but me “This set of properties”. If you look at geolocation data, you see my location every single day after 6:00 PM and now you know that’s my house, you’ve just figured out my address. Now with privacy budgets, it only allows you to pull or query so many of these types of things, so it obviates a little bit of that threat vector, but nonetheless a determined, motivated person could violate your privacy and they don’t need that much data to do it. But the vast majority of people are not out to do that and that threat vector tends to be very specific and very context dependent.

Broadly, you’ve made the argument all along that IDFA is deeply problematic. I think that there are legitimate concerns here, but we went from, “Okay, we have a specific number tied to a specific individual” to basically “We have no linkages whatsoever to anything”. I think the encouraging thing about SKAdNetwork 4.0 is there are linkages. You don’t know necessarily who it’s tied to, but you need some sort of connection between an ad and a purchase to know if it happened for this environment to exist. I think people forget that Apple makes money from apps that people discover via ads so they have a real financial incentive to at least make this work. I’m very curious, it’s hard to tease out, Apple’s services revenue was slowing a bit. This is what I want to know more than anything, how much did Apple feel the pain of ATT in their App Store revenue? They’ll never tell us, but that’s what I want to know more than anything.

ES: And then how much of an imperative now is it to then make up that slow down with ads revenue? Because it’s not just that, I mean that obviously there’s a symbiosis there, but it’s also with the impending breakup of its exclusive access to payments on iOS, it’s going to lose something from that. Now I don’t think that’s going to be that extreme, at least not at first because the native positioning is very valuable, but nonetheless they’re facing some threats to their business there on the services side, which has been the strategically important piece of the company since 2016.

ATT, Recessions, and Shifting Spend

I want to get to the Apple ads bit a bit, but just to back out and think about this: how much of the spend killed by ATT will actually move to say an Amazon or Google or Apple. The issue I’ve written about is that both Google search and Amazon and App Store ads really depend on the user stating what they want and then giving to them, or less charitably harvesting the clicks because the ads is higher up on the results page than the organic result. The issue with ATT though is the ads that are being killed are the ones that show you stuff you didn’t know you wanted. How much of that spend — and you predicted that a lot of the spend is just going to disappear — how much of the spend do you think has really shifted to these more user-directed platforms versus just being out of the game?

ES: I don’t think that much has shifted right now. In that reporting that the Wall Street Journal did about Apple’s discussions they had with Facebook where they wanted to try to monetize some piece of that product offering, I think they estimated the current size of the ad business at $4 billion. If you take Facebook at its word that it lost $10 billion, well right there you’ve got some differential, and there’s more billions than just that. I think the previous assessment on Apple’s ad business was $2 billion, so it roughly doubled from $2 billion to $4 billion, but that’s with no new material placements, that’s just absorbing some of that money that fled from the big ad platforms. I think a lot of that just disappeared, I think it just evaporated.

I think we’ve talked about this before, but the way that performance advertisers they operate, it’s not that this money is just fungible and I’ll just I’ll take it from one place and put it elsewhere, no. It’s that you spend the money across all the platforms and you have ROI standards and as you’re meeting them, you spend, and as you aren’t, you decrease ad spend because there’s an inverse correlation between ad spend and ROI usually. The more you’re spending, the higher your bids are, the more you’re reaching people that are maybe not your core audience, the more you’re reaching people more than once and so the performance degrades. If that ROI drops, you can’t just take that money and allocate it to another platform. First of all, you probably would’ve been working on that platform at the threshold anyway, in the first place.

This is so key. The amount that you spent on ads was not gated by an executive decision. It was gated directly by what your model fed back you were making for that spend. This was the brilliance of these platforms where they have effectively infinite inventory, obviously it’s not completely infinite, and that’s why you would’ve prices go up over time to reach the users that you want and this is why it was so brilliant, particularly in recessions because you would have the brand advertising go away, and sure that would lower prices because there’s less demand for the ads, but the lower prices would draw in more spend from these direct response advertisers, because they don’t care about the recession. All they care about is “Am I getting the ROI on this spend?” and if prices go down my opportunity to buy more increases and yeah, that’s gone, the implication of it happening automatically is that it also goes away automatically at the same time.

ES: It’s almost worse now. I did this presentation about what happens to advertising in a recession? I did this at the beginning of COVID, I thought we were going to enter like a deep depression. We didn’t, nonetheless I made the presentation. If you look at advertising performance during a recession, basically just 2008, the drawdown of ad spend is more severe than the overall GDP, but it recovers faster, it’s more extreme. The thing is, yeah, you have the opportunity to take that brand spend and shift it to direct response cause that’s what you do. You want to have a faster turnaround time on your ad spend between ad spend and revenue. You want to have more certainty around the return on the ad spend and so you just shift it to direct response. Well, now you can’t. So now it’s not only “Are we getting rid of the brand spend?”, because that’s just the first thing you cut in a recessionary environment, but you can’t even reallocate that to direct response because the ability to do so has been degraded.

From a platform perspective, if you’re a gaming company, your addressable market is like zero marginal cost to get one more app installed. So you could just increase your spend, but now that’s not going to be there either.

A Potential Apple Ad Network

Do you think that Apple is going to push ads into apps? And if so, what does that look like?

ES: It looks like what they’ve done in the past. I think they were early with iAd, that market wasn’t that big, it wasn’t big enough to satiate Apple. What do they care about a couple billion dollars?

I think the problem is iAd too was really all about brand marketing. You go back to the early presentations and I remember their canonical one was the Nissan Leaf and it was going to be very beautiful and all these sorts of things, whereas the opportunity is really in direct response, which looks a lot different than that. I just don’t think iAd is really applicable in any meaningful sense other than yes, Apple previously did an app ad network. But everything about it, I think was totally different than what a future one would look like.

ES: Of course. I just mean that they had ads that showed up in publisher apps. But yeah, to your point, it was a pursuing a different strategy there but I think they have a playbook for at least doing that and the same guy is doing it. The other thing is there’s questions around if this time around if you’d even need to have an SDK integrated into your app to show Apple Search ads.

No. The SDK will be integrated into iOS, just like Payments. Like you don’t put an in-app payment SDK in your App.

ES: Exactly. So, that’s a massive leg up.

**That’s how Apple’s going to be able to claim it’s all first party, “We’re using first party data in the ads. It’s not the App maker’s ad, it’s our ads using the same infrastructure as the app store, blah, blah, blah, and we’re using our data. It’s never shared with the App maker. We’re just trying to help App makers succeed.”

ES: Then the bigger question in my mind is, my sense is you don’t need an SDK, you don’t need to integrate an SDK, which already makes this much more appealing than pretty much any other ad network that you’d work with as a publisher. What if Apple Pay data can be harvested in the same way post ad click as IAP data is now.

Oh, so you could start monetizing real world stuff. Yeah.

ES: Right. So now, you’ve got an e-commerce ad network. It’s not just app install ads.

Oh, I didn’t even think about that. Yeah, geez.

ES: Those are pretty big questions. Or Buy-Now-Pay-Later, they’re moving into that space. There’s just a lot of things and when you own the chips and you own the OS there’s just so many things that you get privileged access to, and therefore so many ways for them to expand this. Now, I think they are being more measured about it because of the scrutiny of this whole endeavor.

What do you think about Apple sending around this article about, “Oh, ATT’s not actually really having an impact.” Are we really sure about that?

ES: I don’t know if I want to go down that rabbit hole.

(laughing) You had a good thread on Twitter which I will link too, going beyond my post, saying, “No, this is ridiculous, it’s not true. It’s inaccurate.”

ES: I just feel like if you wanted to look for examples of companies that were affected by ATT, there’s no shortage of them. Look at any of the big public D2C companies and these companies are saying very explicitly, “ATT created a massive headwind for our business.” By the way, people were saying that not in a speculative way but in a very observed, measured way, before anyone was talking about a recession.

Right, and then there’s the fact of the matter, we’ve been talking about this for two years and it’s basically played out, particularly you, you laid out exactly what’s going to happen, and that’s exactly what happened. So to turn around and say, “Oh, maybe it didn’t have any impact” — it’s just weird, especially for Apple to be pushing this. It does feel like, “We’re feeling a little bit of the heat on this and so let’s distract and get away from it.”

Apple’s Alternate Approach

Just to rewind though, you had three things about SKAdNetwork. You had the gradation sort of privacy, you had the faster and set time limits for when they’re giving you feedback, what was the third one?

ES: Just to recapitulate, so there’s hierarchical source identifiers, hierarchical conversion values, the third one is multiple conversions and then there’s a fourth one — those three are the primary considerations for me. There’s a fourth one that SKAdNetwork can now handle attributions from the web, so there’s four big new pieces of functionality coming.

There is one more point I wanted to make about SKAdNetwork: I would say that if at the outset, if at WWDC 2020, what Apple had presented to advertisers was SKAdNetwork 4.0, the spec, and they had unified the personalized ads texts across Apple’s own opt-in prompt and the ATT prompt, and Apple Search Ads used SKAdNetwork instead of the Apple Ads Attribution API, if all those things were true, and those all seem like reasonable things to ask for, I would’ve applauded ATT. I would’ve said, “This is the right thing to do, this is good for consumers, this is the direction that advertisers should have predicted they were going to have to move in.” I would’ve said, “ATT is excellent and I fully support this.”

The problem was SKAdNetwork 2.0 — 2.0 is what they unveiled at WWDC 2020 — was just totally dysfunctional and didn’t work. Then the differences in treatment between their own network and their own opt-in prompts between the ATT prompt and other ad networks. If all those things were addressed, I would be absolutely an advocate for ATT, I think it would’ve been the right policy, but it’s just the preferential treatment in addition to this totally dysfunctional SKAdNetwork, which to your point, was unnecessarily dysfunctional, it created pain that wasn’t necessary to protect consumer privacy.

How much do you think this was malice versus incompetence? By malice, I don’t mean trying to hurt them, but more, “Hey, we’re going to build up our own network.” Maybe self-preference, or was it just the people in Apple that were gung ho about privacy just basically got too much power and overdid it?

ES: That. I think the privacy absolutists won out, they won the day. I’ve heard that there were two camps going into this. There was one that wanted to take a more moderated approach and there was one that wanted to take more of an extreme approach, and the extreme approach won. Then the moderated team made their case after seeing the impact, and that’s why we get SKAdNetwork 4.0. It wasn’t impossible but it was very difficult to foresee what the damage would be.

In fact, I think that’s actually a big risk right now, and I think it’s one of the reasons why I find this idea that ATT hasn’t had any impact, “It’s all macro” or “It’s all TikTok” and these companies are just whining, I think that’s a dangerous line of thought, because I think the primary risk that people aren’t appreciating right now is that actually we’re overestimating the macro impact and we’re underestimating the ATT impact.

There’s actually compounded-over-time impact that we will only see play out over the next year or two years. Where I was spending some amount of money on direct response ads, I was making my money back in ninety days, I was taking that money and I was reinvesting it, and I was making that money back in ninety days and I was making a 10% premium, and I was taking all that and I was reinvesting it, and I made that money back in ninety days. So I’m compounding my revenue, I’m growing my revenue via that compounding because every time I get that 10% margin on ninety days worth of ad spend, I reinvest it and so that 110% grows by 110% and on and on and on. Well now, if I’m taking that and I’m shrinking, and I’m going to be shrinking it every ninety days because the ads are less performant, the ad platforms can do less to generate revenue.

So that headwind or that sort of counterfactual is not just that first quarter of ad spend where I’m spending less, it’s every subsequent quarter, and that’s a continuous drag on growth going forward, and my worry here is that we’re under-appreciating that risk.

Yeah, I think that’s a great point and something to watch super closely is once the economy turns, what was actually the real impact?

Unity, ironSource, and AppLovin

Let’s talk about Unity. Obviously, it’s been quite a year for Unity to say the least. I thought the reaction to the ironSource acquisition was pretty interesting. You and I both came out right away with, “Oh, this makes a ton of sense, this marks a very clear way forward.” Everyone who thinks of Unity as a game engine, on the other hand, was appalled, and so it’s kind of weird seeing the dichotomy and reactions. Walk me through why this acquisition made sense and then we can get to the AppLovin part, but I think it’s good for people to understand how this stuff works, how it’s tied into ATT, all relevant to the stuff we’ve been talking about.

ES: I guess as of today, the AppLovin proposal is no longer a real live consideration, but we’ll see how that plays out, I don’t don’t know.

So Unity and ironSource — I think that the combination did make a lot of sense. Basically, you had the number two and number three mobile game app install ad networks. Unity has been late with a lot of innovation to my mind and so they only just launched a mediation platform, a monetization platform in October, but ironSource had one since 2019, a scaled mediation platform.

And mediation is basically where you can get bids from multiple ad networks and then choose the best one to put in, so instead of you being an app maker having to install SDKs from every single ad network — it’s called header bidding — but you take the head of the snake, and they will put it out for a bid and then you’ll just serve the one that will theoretically make you the most money.

ES: Right, exactly. So these mediation platforms, really just supply side platforms (SSP’s), they aggregate demand. So you’re getting a lot of demand for your inventory, then more competitions should drive the prices up.

And Unity — was it like a strategy tax? Because they have such a large ad network, it’s like, “Well, people will just put our ad network in directly, why would we want to have a product that basically services other ad networks?” Was that why they were slow or does that not make sense?

ES: It does make sense, the evolution of this category is kind of interesting. That definitely was the approach a couple of years ago, so when mediation first surfaced in this space, it wasn’t done with real time bidding, it was done with this waterfall method, so you’d have historical CPMs for the different ad networks and you just prioritize them, which ad network got which impression, and then what ordering, and you would query them when an impression became available, based on just historical performance of that ad network, so you’re leaving money on the table because it’s not real time. This ad network may have performed really well last month but it’s not performing that well anymore, and so it’s not ideal. And so the introduction of header bidding, really in-app bidding, is what it’s called because there’s no header, the header pertains to the JavaScript header for a website.

So within app bidding, there is a real time auction for every single impression, or that’s how it’s supposed to be run, but in actuality, a lot of publishers use a hybrid approach. So they’ll have a waterfall set up, but certain parts of the waterfall are in app bidding and some are just queried against the ad network. So when mediation first surfaced, it was that waterfall method, and what happened was a lot of the ad networks saw that as a threat, and they said, “Okay, well let’s just do all these exclusive deals with people. So we’ll exclusively service your inventory, we’ll pay you a fixed CPM or we’ll have a CPM floor and you don’t need to worry about this waterfall method.”

As mediation evolved to include more in-app bidding, to include the real time auction, then that approach just didn’t work because publishers said, “Well, no, I’d be leaving too much money on the table. I’d rather just have every single impression be competed for.” The only problem with that was that not every single demand partner wanted to participate in the real-time auction, and the best example of that was Google. They didn’t want to bring their demand into the real-time auction, they only wanted to do that if you used their mediation platform which is called AdMob.

It was this kind of messy system, it was this hybrid set up, it was kind of complex. If you look at any sort of ad monetization manager’s dashboard, you’d see this massive web of logic that pertained to the different tiers on the waterfall. Well now, Google announced deals with AppLovin and with Fyber and with Chartboost to all run SSPs, and they said, “Okay, we actually will make our demand available in the in-app bidding environment.” So that was the last barrier to be broken down that enables basically full in-app bidding.

Right and so now it’s the obvious way to go for every developer, and Unity’s stuck with being years behind, and so they needed to acquire one, that’s basically the long and short of it.

ES: Right, but I think in the post ATT environment, owning supply becomes even more crucial because the data that you get in serving the impressions is more voluminous than you’d get with filling the impressions. So if I fill an impression, I get one data point. I get if this bid won or not, and in some cases, I get the price of the bid that did win.

So basically you need to be able to see every single piece of inventory across a whole universe of apps, understand which ads are working, which ads are not, where they’re coming from, what the prices are and you’re basically backwards engineering the model that you had before, and you’re intuiting this particular user or this particular ad at this particular price is successful, I’m going to push more in that direction.

ES: Right, you’re just getting much more price discovery data in a single auction because you’re getting all the bids that come in. What is a bid? It’s a proxy for the expected value of buying that impression for that advertiser, so you get a lot of knowledge of how their apps monetize, that’s the data that feeds that calculus. Now on the demand side, which is how you make your money, which is, “Okay, this is how much you need to bid to win that impression”.

Unity owning a mediation tool ought to make the Unity ad network more profitable because they’ll better know how much to bid for different inventory.

ES: Right, and that’s exactly what AppLovin did in buying MoPub, which was the largest scaled independent SSP. Then what did they announce almost immediately after doing that? They were going to divest some of their games, and so my theory there is, well, if you want a bunch of game data, game monetization data, and you want to get that from your own proprietary games, you’ve actually got to buy it, you’ve got to go buy users to come in the games to generate the data. Well if you run an SSP people will pay you to give you that data and so maybe you’re getting the same sort of quality of data roughly and you don’t need the games anymore because you’re getting it on the SSP side.

How is this impacted by ATT, if at all? It sounds somewhat sketchy like you’re going to have these SSPs and all these different apps, all these apps are third party apps. Again, you’re moving away from the AppLovin model of it being first party. Is this just one of those things because it’s all server side, there’s nothing that Apple can do about it?

ES: Well, this wouldn’t violate ATT because this is just price data, like estimated value. It’s not user level —

There’s no user data at all.

ES: Exactly, it’s not tethered to an IDFA. It’s all contextual, this is a contextual target, and the thing is these companies for the most part they were doing primarily contextual targeting, but they also had device graphs and they lost those. Why wouldn’t you? I mean that there wasn’t anything wrong with that, but you knew which IDFA installed which games, and that was knowledge that helped you in the demand fulfillment process.

Yeah, that’s a great point. So why did AppLovin respond by trying to buy Unity? Is this just a matter of was it the Unity ad network that was valuable? Was it the Unity ability to get in the front door with developers? Because we’ve already used the Unity engine, just add on the Unity SSP. If it’s there, are they worried about getting shut out of this market by Unity plus ironSource? What was the thinking in moving to this direction? And then, yeah to your point, Unity just turned them down today. Are they going to come back? Why would Unity not want to go with AppLovin, given they are larger than ironSource?

ES: The AppLovin proposal didn’t value Unity, I think, in the way that Unity believes it should be valued. Also AppLovin has pretty considerable debt load, they’ve bought a lot of top line revenue growth. AppLovin shareholders would’ve owned the majority of the company, John Riccitiello would’ve been CEO, but AppLovin shareholders would’ve owned the majority. So I think there were a lot of aspects of the deal that probably weren’t ideal, versus the ironSource deal to my mind makes a lot of sense. So you’ve got a company that was EBIT positive, very well respected, very well run.

Very well respected in the narrow ad industry. They’re still carrying the weight of their installware business model from like a decade ago.

ES: Yeah, that’s true. So I went out and I was in Israel one time and the ironSource people were like, “Hey, you’re in Israel, come hang out, come to our office for lunch.” Saw them for lunch, and they were telling me about that company, and I was like, “Wow”, I had no idea they were involved in multiple lines of business. I always thought that they were a mobile game app install ad network, but they had a desktop business and all this stuff. My understanding is that they jettisoned all of that.

Yeah, they did. Actually I think it was actually tied to the building of their mediation product was a pivot away from that business. They shed all that and moved towards building this new thing, which is what actually makes them valuable now.

ES: Right, this is the company now. I learned about that in the Twitter ruckus, and according to Wikipedia, they don’t run where that product was followed anymore, I don’t know too much more about it. But in any case yes, as a mobile game in-app install ad network, they’re very well respected. Strong operators. Tel Aviv is the Mecca of Ad Tech and so I think in combination, it’s a really strong company, and you can understand why AppLovin would be intimidated by that.

Yeah, it makes sense in that regard.

Meta and the Benefit of Scale

You mentioned this a little bit earlier, but as we’re wrapping up here, looking forward you can see Apple maybe doing some more crackdowns on fingerprinting or some of this third party stuff, there’s potential regulation. It kind of feels inescapable to me that there’s going to be more consolidation. The returns to being very large are going to be huge and I think particularly for Meta, it’s pretty interesting because if you have this situation where 1) you need to be large and 2) ads need to be increasingly contextual, which is going to just require a lot more computing capacity and modeling and things along those lines relative to an IDFA, which was kind of straightforward — you just match column A to column B. Is it wrong to actually be somewhat optimistic about Meta’s fortunes looking forward?

ES: No, I think you’re exactly right. So Sheryl [Sandberg] — what a run she had at Meta — in her last earnings call, brought up click-to-message ads like five times or ten times.

We’ll miss the Sheryl anecdotes in general, I should note.

ES: Yeah, she seems very optimistic about those and that kind of aligns with my content fortress thesis. [CEO] Mark [Zuckerberg] even said that one of the big initiatives of the company is to grow first party understanding of people’s interests by making it easier for people to engage with businesses in our own apps. So just to generate much more first party data that they can use to target ads. The other thing is Reels, obviously if your ability to target ads has been degraded, the value of the ads decrease. Well, how do you maintain the same amount of revenue? You show more ads, and you probably don’t want to increase the ad load — ad load is the number of ads seen per session. So then you want to increase the session length, or per minute of session or whatever is ad load and so if you increase the session length, then ad load can stay the same, you get more ads. Even though they’re lower CPM, which is exactly what happened to them last quarter, they had total ad impressions was up 15%, average price for ad was down 14% and they were down 1% year over year on revenue.

This is the counterintuitive thing about these businesses, that’s great news! I remember back when Google first IPOed and people would look at, “Oh, the cost per ad is going down.” It’s like, “That’s good!” It’s very counterintuitive because it’s dangerous when — this happened to Facebook a couple times — where they just run out of places to grow as far as ads goes.

This happened right before Stories. Their price per ad was going up, and people were getting excited about this, “Wow, look at the pricing power.” And it was like, “No”, this is actually a very problematic sign because that means they don’t have sufficient inventory to show, that also means advertisers are going to other platforms, it’s giving other platforms money and the ability to increase their competitive product, and Stories was this huge triumph from a business perspective because they just exploded their ad inventory and their price-per-ad plummeted.

It’s hilarious, because their stock got killed, the second biggest drop after Facebook’s first greatest drop this year was Facebook after the Stories earnings, when they announced that this was happening. I remember at the time I’m like, “No, this is really great news,” because it basically sets them up for multiple years of growth opportunities, which is what happened. And just to double down on your Reels point, them focusing on Reels, it’s kind of like of all the stuff that’s gone wrong for Meta/Facebook, the fact that they’re doing this Reels double down at the same time they’re having these ad business challenges is actually a piece of good fortune in my opinion.

ES: Yeah, I think so. Sheryl even mentioned that this is the playbook. It’s like we get a toe hold, we find a new front and we expand our inventory and then we get better at targeting ads in that inventory, so the value of the ads goes up. But that’s a process, you have to find the new port point of entry for showing these ads and then you’ll over time increase the value of them because you’ll get better at understanding what signals to use for targeting.

Just to follow up on what you said, it’s important, I think we’re getting to a point where size, scale, sophistication, all of those things are going to be really important just for any sort of competitive staying power, and especially with SKAdNetwork 4.0. Because 2.0 and 3.0 kind of leveled the playing field.

(laughing)It was terrible for everyone.

ES: Yes, exactly. With 4.0 your ability to parse out meaning from the added context that you’ll get with potentially three postbacks with the hierarchical conversion values, Facebook’s going to be able to do a lot more with that than probably any other ad platform except for maybe Google. So they’ll be better positioned to take that new context and build value out of it. If you’re thinking about — and Sheryl said this too — we’re at the beginning stages of clawing back the measurement that we lost. But I think they’re very realistic about where they are in this process. Just giving them more surface area to extract value from will benefit them in a way that a lot of the app install networks probably can’t do. Maybe even like a Snap, a Pinterest, a Twitter can’t do, I think it’s going to make the topography more diverse again so the playing field will no longer be level.

Yeah, you just need rules of the road. I think the problem with SKAdNetwork 2.0 is because it was worthless, everyone was just adrift on the sea and with the new SKAdNetwork, if you actually know the rules to play by, yeah those rules might be crappier than the rules you had previously so you’re going to have to build it again, but as long as there’s something you can rely on, and something you can model against, you can build something again. But to the degree it’s hard to do, the more that favors Facebook. It’s kind of like the GDPR thing, I wrote this at the beginning, I’m like, “This is so clearly going to benefit the biggest companies because they can afford to hire all the compliance officers, they can afford to do all this work.” I think it’s going to be a similar dynamic here.

ES: 100% and I think it’s doubly true with the AI initiatives. I think when Zuckerberg talks about AI, he doesn’t mean it in the way that most people interpret it. He’s talking about parsing signal out of video.

And it’s going to be the same problem. Show the right video to someone in Reels, and show them the right ad, it’s the same challenge.

ES: Exactly. But it’s a much more appreciable challenge than doing that with text, and who can compete there? Facebook’s got its own data centers with its own hardware, that was always Snap’s problem, by the way, their cost of revenue was so high because they were totally dependent on Google Cloud and Amazon Cloud.

That’s a great point. That’s going to actually get worse for them now that they have to start doing this AI analysis of video.

ES: Yeah, Facebook said they’re doubling or tripling a number of GPUs that they’re going to have in their own proprietary hardware, in their own owned data centers. Who can compete with that?

Yep, that’s spot on. Well, I think it’s safe to say we will probably check in after next quarter, but it was good talking. It’s fascinating, there’s definitely no shortage of content opportunities for us, so I guess we should be grateful to Apple.

ES: You’ve kept my keyboard in constant use, Apple.

---

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!