There are a whole bunch of advertising-related topics I wanted to cover after this quarter’s earnings; I figured the most efficient and interesting way to do so would be to bring Eric Seufert back for another Stratechery Interview.
Seufert is the creator of Mobile Dev Memo, where he writes about mobile advertising. Seufert speaks from experience: he worked for several mobile gaming startups when the app ecosystem was just taking off, and was later the Vice President of User Acquisition for Rovio, the developer of Angry Birds. Seufert then transitioned into building platforms for advertisers, and now works as a consultant in addition to his work at Mobile Dev Memo. Seufert also just launched a new fund focused on investments in startups in the gaming, adtech, social, and creator spaces.
My first interview with Seufert was in February 2021, just before Apple’s App Tracking Transparency (ATT) policy came into force; our second was last fall as the impact just started to be felt. Six months later and the carnage is much more visible, and measurable, so it felt like a great time to have Seufert on again.
This was a pretty long interview; there is an index of topics below, and as always, I recommend listening via your member-only podcast feed.
On to the interview:
An Interview with Eric Seufert About the Post-ATT Landscape
This interview is lightly edited for clarity.
Impact of ATT | The Post-ATT Landscape | Measurement and Targeting | Google and YouTube | Apple’s Ad Network | Shopify Audiences | Regulation and the Future | The Ad Agency Opportunity
Impact of ATT
Eric Seufert, it’s good to talk to you again. The last time we talked was six months ago and I titled the interview, “The Impact of ATT”, it turns out we were a bit early because this was after the Q3 earnings last year. It was really the next quarter when Facebook in particular was hit really hard and then this past quarter, it’s hit everyone. It feels like this was a case where you and I were too in the weeds, where we saw it playing out and we said, “Yup, there it is. ATT’s coming,” but it felt like the impact that actually hit the financial results for the market to wake up. Is that the case or was the impact actually even worse than you expected?
Eric Seufert: My theory here is that Q1 2022 is where we saw for the first time the undiluted impact of ATT. There were always confounding factors in the last few quarters. Even last quarter, when we saw the stock price reaction to ATT, that was very much based on —
Looking forward, forecast.
ES: Right, the forward looking estimate of the impact. I think Q1 is where we saw the real undiluted impact, and why do I think that? ATT gets introduced as a mandatory required feature of iOS app developers in April with iOS 14.5. iOS 14.5 never reached more than something like 20, 25% of devices. 14.6 was rolled out and again, hovered at that 25% penetration level because Apple controls adoption of new iOS releases with the push, the notification you see on the Settings app. And they didn’t do that, they didn’t do that until just before WWDC. So you didn’t get ATT on a majority of iOS devices until July. So, there goes Q2, Q2 is basically unaffected. Q3 was the first quarter where ATT is in the market on the majority of devices.
And that’s when we talked was after that quarter, like “It’s definitely starting to happen here”.
ES: Well, advertisers saw the impact, but there’s a couple pieces here, because I get this question a lot. On the advertiser’s side, especially if you look at app advertisers and especially the categories that I thought would be most impaired by these is very core games. DTC brands saw immediate impact, but on the gaming side, these core games, on any given day, something like 95% of the revenue generated by the game would be contributed from users that are more than a month old in that game. These are cohorts that tend to be very sticky once they reach some point of retention, some retention threshold. Just not replacing them is the impact of ATT, those cohorts are there, they’re not going to stop spending money.
There are people that are spending money in a game and they’ll spend a money in that game for years and that props up a game for a long time so you don’t see the impact of failing to add new users. It takes a while for that build up.
ES: Exactly right, there’s a cumulative effect, but that’s on the advertiser side. On the ad platform side, those algorithms were seeded with data. So let’s say the very first day that ATT is on a majority of devices. Now, let’s say you’re Facebook, you’re Snap, you’re TikTok, first of all, you’ve still got, let’s say, 15 to 20% of people opting in, so you still have a baseline of data to sample from, and then you have the algorithm seeded with data. These algorithms work on lookback windows — it could be 90 days, it could be 180 days or whatever — but they have a lookback window and so you’re losing one day from whatever that lookback window is. Let’s say the lookback window is 90 days, you’re losing 1/90th of the data, which sees the algorithm, which tunes the algorithm. So Q3, you had a period where the parameters of the algorithm were degrading, but they didn’t go to zero. They don’t immediately go to zero just because ATT all of a sudden is on a majority of devices. They bleed out, they just degrade over time. So, that’s Q3 and then Q4. So over the course of Q3, that was happening. The algorithms were losing their capacity to target, but on a day-by-day basis, because of this lookback window effect, and then Q4 is just always messy because brands dump budget. Q4 is Q4, Q4 is holidays, it’s brand spend and a lot of direct response advertisers pull back anyway.
I want to double down on that, because I feel like these dynamics about the market are so important and it goes over so many people’s heads. What you just said is you have Q4, which is a big brand advertising period for obvious reasons, seasonality, it’s the holiday season. But because all these platforms are auction based, if the brands are coming in and gobbling up inventory, that by definition is increasing the demand, which is going to increase the price. Many of these buyers, particularly for the games, they’re so finely tuned, they will not buy ads that don’t have the right ROI on them, return on advertising spend, ROAS. So when brand advertising comes in, they always pull back. Again, I know you know this, I’m just re-emphasizing this because I think this bit about how sensitive these markets are is not just important for this discussion, it’s important for the discussion broadly. I just wanted to double down on that point, but sorry, I interrupted you. Please continue.
ES: But it’s an important point. It’s an important point because first of all, I want to make the case that this is not a game-specific phenomenon. This is not a phenomenon that’s unique to games, because sometimes I think people interpret it this way and they’re like, “Well, yeah, games are games. It’s not the entirety of the market.” This is e-commerce, they’re just as precision-tuned, they’re looking at day-to-day average order value, looking at day-to-day average ROAS on a day zero basis. “We bought this cohort of users today, how much of the expenditure for that cohort did we recoup in the same day?” That informs our bid decisions tomorrow or today so this is like a real-time tuning exercise, this happens in real-time. If you say, “Hey, our day zero ROAS is not hitting our threshold, the threshold that our model tells us it has to in order to recoup our spend by whatever timeline we’ve set as our standard,” then we pull back spend, we reduce bids. Well, we reduce bids and that results in spend decreasing. This is very much a real time exercise and is informed by a vast amount of data, but also, the modeling that goes on within these advertisers data science teams and marketing science teams to inform that process, this is not blind speculation. This is a very scientific exercise that happens in real time.
We’ve moved far away from the famous saying, “Half my advertising works. I just don’t know which half.” The benefit of this whole system is you actually could know how well your advertising was worth and that gave such a leg up to — I mean, yeah, you could afford to waste half of your advertising if you’re the dominant CPG player in the country — it’s a lot different if you’re a small retailer or developer. You need to know how that’s being spent, and if you can that creates all sorts of new opportunities.
ES: Exactly, and then you look at the advertising base of Facebook, and that was small-and-medium sized businesses (SMBs), they talked about 10 million advertisers, well the vast majority of that revenue base is SMBs. Obviously, the spectrum of spend was very long tail and you had people spending vast amounts of money, but the bulk of the revenue was from these SMB, and I think that argument because Facebook made it…
I know. It got ignored.
ES: It got dismissed just out of hand. It’s prima facie false because Facebook said it, right? Because if Facebook says something, obviously, it’s a dishonest argument or it’s a manipulation or it’s an untruth — but in this case it just so happened to be the truth. These small companies that do make these day-to-day, real time decisions about bids, which then influence their spend amounts, they are small companies. A DTC brand could be like five people, and the CEO is doing the marketing, the CEO is in Facebook Ads Manager or Snap Ads Manager, and so you saw that very immediate reactivity in Q3 and in some cases, it was overly aggressive because people were worried.
The thing is, it’s not so much that the performance is going to decrease, everyone assumed that was going to happen, it was that your model breaks, because your model is developed on a traffic base that is going to look markedly different post-ATT than pre-ATT, because the users that were being targeted look different. They have different profiles, because again, you’re not able to do that user level targeting anymore to get hyper relevant subgroups of people, you’re getting much coarser traffic, you’re getting much broader traffic. So if your app or your DTC products or your e-commerce store was very dependent on getting those very relevant people that are primed to purchase things that are the things that you sell, then the composition of your traffic is going to change fundamentally, and so you need to just go to zero and rebuild the model from zero, because your model was predicated on a user looking a certain way and having a certain level of relevance for your product. If that changes, then the model is no longer relevant. Whatever the model tells you is probably flawed because it’s making assumptions about the relevancy of the users and the composition of the traffic base is just no longer true post-ATT.
Right, because you’re assuming that you’re spending a lot of money on this customer acquisition, you’re doing it because you’re confident that if I spend $5 on this user, it’s going to deliver me $7 in margin or whatever it might be. Obviously that nets out to two or whatever the number it might be, but as soon as you lose sight of that, you’re really running the risk of just instantly bankrupting yourself. If you lose a handle on that ROI, then you could get yourself in trouble really quickly.
ES: Exactly, because the payback windows in a lot of cases are multiple months and in some cases, could be a year or longer. When you’re buying, you’re making buying decisions on the basis of that day zero return on ad spend but that day zero return on ad spend is an input to a model that tells you, “Okay, in the next 90 days-”
What’s the lifetime value of this user?
ES: Exactly, and then that’s based on the economic profile of that cohort like how often do they spend, how much more are they going to spend after day zero. You could get people that come in and the day zero looks really bad, but actually, the lifetime value is going to be much, much worse on a relative basis than the day zero looks because they’re not going to spend anything subsequently.
Yeah, you might be a user where you get a bit on one thing, but then you have their email address, you can direct market to them going forward and you’ll get future purchases from them.
So, we’re through the holiday quarter there, we had a little bit of a diversion explaining the differences. So, then just this last quarter, basically, all those excuses or all the data is now fresh, it’s all post-ATT, the holiday’s over and boom, it’s super clear that this is a big impact.
ES: Exactly. The algorithms have withered, all of the pre-ATT data has flushed itself out of the algorithm via that fixed lookback window, and so the smoke has cleared and you are left with an unperturbed view of digital marketing post-ATT.
The Post-ATT Landscape
What’s that view? What is the landscape? Is it better or worse than you expected?
ES: It’s worse. I put together an analysis of what I thought the impact on Facebook’s business would be, I focused on Facebook because that’s the advertising platform I’m most familiar with as an operator. I had a base case, best case, and worst case. In the base case, I said, I think it was 7% revenue headwind, and best case was like 3% and worst case was like 12%. It ended up being about 12 to 15% revenue headwind, so it was worse than I thought. I think we talked about this last time or maybe it was the first time that what are the IDFA or what is device, they’re called MAIDs, mobile advertising IDs.
I blame the advertising industry for overuse of acronyms. So it’s great to see you get tripped up by it now, but continue.
ES: (laughing) Right. Mobile advertising IDs, because we’re not just talking about IDFA anymore, right? This conversation has expanded to GAID, Google Advertising ID, because Google announced that, I’m sure we’ll talk about that in a second. But anyway, abstract this to user identity.
When you know who the user is and you can trace that user from an ad interaction, ad view, ad click through to whatever economic impact they have in your product, the entire user lifetime is from the first moment they interact with your ad through to when they churn. That level of transparency and that window into their actions points in a way that bridges all the ad engagement metrics with the in-product metrics, that just gives you a full user lifecycle view, and that allows you to make those determinations to know that this click resulted in that. Again, this is under a last click model, which some people might argue is antiquated or insufficient, but in any case, it gives you the full picture of the ad interaction cost, and then the in-product’s monetization benefit. It allows you to do that accounting, simple accounting — this is how much I spent, this is how much I got as a result of that spend.
That breaks with ATT, and then everything that it allowed subsequent to that, that creates the ability for a feedback loop of saying like, “Well, when I know that whatever I did with the ad resulted in high margin, the ability to spend a lot of money and profit, I should do more of that and whenever it didn’t result in that, I should do less.” So the ability to make that decision and the ability to make that assessment then allows you to do more of the stuff that works and do less of the stuff that doesn’t, and that’s targeting. That’s changing your ad creative, that’s targeting different groups of users, that’s giving the ad platform the agency to go and find lookalikes, users that look like the ones that monetize, that are valuable, that are high value. You lose that targeting capability without the MAID, without the user level identity construct.
But in the case of ATT, SKAdNetwork is dysfunctional. There are a lot of components to this that just break that operating model so fundamentally such that you can’t even do measurement. I don’t even know when I spend a dollar, how much money am I getting back? That’s the real problem that these platforms have been grappling with and they’ve taken different approaches to solving it.
However, I wrote a Twitter thread about this post-Facebook earnings and I said, “Look, you have to think about ATT like a plane crash. Now all these advertisers are adrift in the ocean and they’re in a rubber dinghy and the goal is we have got to get to land”. That’s the long term goal. We want to get the land, but actually, the dinghy’s got a hole in it, and the short term goal is if we don’t fix this hole in the dinghy, we’re going to drown. That’s the measurement gap right now. It’s just having the ability to tell an advertiser, you spent $1, you got a $1.10, or you spent $1 and you got 90 cents and they’re all trying to fix that. They’ve made varying degrees of progress against that. Facebook announced a couple months ago that they believed that they were under-reporting conversions by 15%. So if there were 100 conversions generated from a campaign, they could account for 85 of them, and then they published this blog post saying, “Well, we’ve actually increased that to 92%.” So now if there’s 100 conversions, we’re able to account for 92 of them, but there’s still that gap of 8% and that’s a problem.
Forget about the ability to do targeting, that’s gone or at least to the extent that they were able to do it before, that’s gone forever. When you go from individual level targeting to group level targeting, you just lose some precision and that’s gone forever. Whatever that delta is, that’s gone forever, but the measurement piece probably can be fixed. That’s just a modeling issue, that’s a data science issue and Facebook has made progress on that.
Measurement and Targeting
How are they doing that? How are they doing measurements in this new world?
ES: I wish I had a reliable answer. My sense is it’s the Conversions API (CAPI), they launched this Conversions API in 2019 basically as a reaction to Apple’s ITP policy, which is like ATT for the browser. One thing I think that’s interesting to note here is that Apple tends to launch these privacy initiatives on Safari first and then there’s usually a one-year gap and then they apply whatever that was, if it was successful, to the mobile apps ecosystem.
So the fact that there are disguising IPs on Safari is bad news for apps.
ES: Yeah, I believe so. I mean, you could make that assumption.
Walk me through the Conversions API.
ES: ITP basically broke third-party cookies. Facebook made its cookie first-party as a result. Apple changed ITP to impose some further restrictions on first-party cookies and then Facebook said, “Okay, this is a game of whack-a-mole. We’re going to skip to the end.” They created CAPI. CAPI is a total server side API for sending conversion data to Facebook, so it’s totally outside.
It’s not going through the browser at all. It’s just going server-to-server.
ES: Exactly right.
So the sense here is that basically games or e-commerce sites or whatever, they have a direct connection to Facebook and they’re sending that this person converted. The modeling is basically Facebook figures out a way this person converted from this IP address at this time of day in this location and Facebook’s like, “Well, we serve this ad to this IP address at this time of day at this location.” and they’re putting the pieces together. Is that broadly correct?
ES: That’s my assumption. My understanding is that they don’t use IP address for that, it’s just basically a signal, but I think you could get pretty precise with that.
First of all, I just want to make the point CAPI was launched, CAPI was basically a tool for Shopify retailers, and Shopify had a one-click integration of CAPI. If you’re a Shopify retailer, you clicked a button when you set up your store and Shopify handled all that, but it was only for web originally. It took identifiers as the indexed identity component for the conversion so that was fully anchored to an identity.
Now, with ATT, they can’t do that, and so what they did was they opened up CAPI to app advertisers. Originally, it was just for web advertisers and I think primarily the point was to service Shopify retailers, but now they opened it up to app advertisers and now you can send all this conversion signal back to Facebook and then they can absorb that into the model, and then try to credit the campaigns that they believe were responsible for those conversions. They’ve managed to claw back some of that measurement gap, not 100%, but I think at some point, they’ll get to 100%.
Snap said the same thing, Advanced Conversions, we’ve improved it, I think Advanced Conversions takes a different approach, we could talk about that. My understanding of that’s based on the Financial Times’ reporting, but if that reporting is correct, then they basically use some of these MMPs, these ad tech middlemen to manage some of that for them and the Financial Times article indicated or alleged that they use the IP address. So, maybe there is more of a risk.
A little bit of a shakier spot, yeah. So with the old MAID model where you knew the identity, it wasn’t just that you could measure the conversion, but then you had a very clear and distinct signal that you could use for targeting. Say you want to do a lookalike audience, right? This person converted, we know for sure, find me a person that looks like this person, because their likelihood of converting is going to be higher than another person. To the extent that this modeling approach where, “Okay, we know this conversion happened, this ad happened. We can get a pretty good sense of what happened.” Is it as perfect as before? Not quite, but why can that not also be used for targeting or can it? I can understand the IP address problem. If you’re doing “fingerprinting,” which by and large means collecting IP addresses. Well, IP addresses are not necessarily long lived, you might be switching devices, you’re switching networks and so it’s hard to find that person again, but if you’re doing the Facebook approach, you’re not necessarily using IP addresses. Is there a path here to refining this approach such that it can be used for targeting? Again, it’s never going to be as good, but is there a path there?
ES: There’s two pieces to that. I would say no, because one is that just modeling that you believe this conversion belongs to this campaign doesn’t give you any insight into who —
Right, you just knew that ad was being shown. You can’t tie it to who it was being shown to.
ES: Exactly. And that’s what you need to use to feed back into this targeting mechanism. Then the other piece is just the targeting has gotten much more coarse, so now targeting is all about groups, big groups, geography, any interests that have historically not been very reliable for targeting, gender, age.
It is very ironic that all the ways that people thought Facebook advertising worked before is actually what Facebook is now being forced into offering. Whereas before it was just like, “Did someone buy something or not?” It was pretty straightforward. You didn’t really care who they were, what they looked like. None of this demographic stuff that gets people feeling creepy about advertising, none of that actually mattered at all and now it’s going to matter a great deal. It’s like everyone had it totally backwards.
ES: Right, exactly. And the thing is in my experience, you would never be able to predict a high value user on the basis of these demographic features. You just wouldn’t and there’s not that much commonality. They almost look like just totally randomly selected because there’s very much a preference issue, there’s very much a disposable income issue, there’s a “just in the moment I feel like spending money right now” issue. Those are not things that you can find commonality around, they’re hidden from the target.
Well, the only signal that matters is have you converted previously. That’s basically it.
ES: Right, exactly. I wrote a piece about this two months ago. I made this point, it’s like a metaphor, imagine you go into the mall and it’s like a random mall. I think I said a random mall in the Midwest because I didn’t want people to be like, “Well, a mall in Palo Alto…” Random mall in Madison, Wisconsin, be both lived there, and you go to the information desk and the person at the information desk says, “Hey, here’s a fun fact for you. The average net worth of people in this mall is $100 million.” And you’re like, “Wow, everyone here is pretty rich.” No, there’s a billionaire there, the odds of everyone in there being worth $100 million or in some range of that, it’s impossibly low. The more probable formulation there is that there’s a billionaire in the mall and everyone else is just worth what the average American is worth; I got that from a [Nassim] Taleb lecture, but that’s what you have to think about.
The entirety of the advertising ecosystem and especially when you get into apps in the freemium economy, but even e-commerce stuff, it’s all driven by fat tails, by long tails. The underlying distribution of the behaviors, it’s all driven by long tails, and the only way to monetize on the product level, but especially with the advertising level because you’re spending money up front to recoup, is to be able to capture those high value people. Sometimes they’re called whales, but those high value people that want to spend a lot of money, and you offer them the ability to spend a lot of money, the only way to make the economics work is capture them, because most of the time people spend zero. A lot of your impressions are wasted, because people just aren’t ready to spend money. So the only way to make these model work is a bottoms-up approach. “I’m going to find those people specifically, I’m going to find those individuals” and when you have to go to group level, there’s just so much loss, so much inefficiency, that the economics can become impossible.
Now, not for every product category, not for every product, but for a lot of products. A lot of the products that were successful because of these tools that these ad platforms created and made available to find those individual people that wanted the opportunity to interact with these products. When that goes away, those products, those product categories in some cases are no longer viable.
Yeah. This is why, I mean, it’s frustrating. There’s literally no winners here except for Apple, which we’ll get to in a moment, especially their future ad network. I mean the people that want to buy stuff can’t buy it, the people who want to use stuff for free are going to have fewer options. Anyhow, we’re on the same page so we can find someone else to argue with us.
Google and YouTube
I want to get to Apple in a minute, but I want to talk through Google and three parts of Google. So, number one is the GAID and they said, “Yeah, it’s going away, but it’s a little fuzzy”, and Facebook’s like, “Yeah, we support their approach.” So that’s number one. Number two is the degree to which Google has benefited or not from ATT. I think this has been particularly interesting in Google Search and e-commerce where it feels like they’re collecting a lot of last click attribution in part because they can measure it and spend is going there and going away from Facebook. Then number three is YouTube, which it feels like YouTube has gone under the radar, but YouTube’s been hit pretty hard, I think, by ATT and they had a much bigger direct response business than each part of Google. I gave you three options there, which ones should we hit?
ES: Let’s move in reverse order, so YouTube. Google earnings and especially the YouTube performance was what made me think, “Okay, this is the first time we’re seeing with clear eyes the impact of ATT.” Because in Q4, YouTube went gangbusters, brand spend, it’s a bonanza, that’s great.
They did say on the call that direct response was not as good, but to your point, that could have been crowded out by brand spend. But Q1, it was definitely not crowded out. It was just bad.
ES: Yeah. So, Q1 was like, “Look, there was weak performance with direct response.” The way I think about the ecosystem is Google is a search company, and YouTube is the display network. They actually have the Google display network, which is a separate thing, but YouTube is their display channel, the big one. That’s where a lot of the e-commerce, the direct response advertisers are trying to get their ad spend there. On mobile, it’s a little bit fuzzy because they just have one product and it’s a black box, it’s called UAC. That was the bellwether for me, it’s like, okay, if YouTube’s having problems with direct response, that means that we’re seeing the real impact of ATT here in a way that you can’t really explain away with the other factors. That’s what they announced in this earnings that they had soft DR spend in YouTube and now the thing is YouTube, my sense would be that that probably skews more heavily brand. My sense would be that’s probably more like a Snapchat profile versus a Facebook profile-
Which is about 50/50.
ES: Yeah, just given the format and how closely it resembles TV. So, when they called that out, that was specific to direct response, the softness, but nonetheless, even YouTube was not immune. The weird thing with quarter two is I don’t know how much I buy that Ukraine created any friction for these-
If you actually read through what they said, it was totally contradictory about what impacted and what didn’t. So, yeah, I thought Google’s call, they were obfuscating a lot, it felt like.
ES: They were talking out on both sides of their mouth. They said, on the one hand, brand was really strong in this quarter. On the other hand, Ukraine caused brand advertisers to pull back and you should interpret our results that way. Well, what, why? I just don’t buy that. Why that would even be the case?
What I’m curious though about that call though, them going back and forth was they’re like, “Oh, search was really strong and e-commerce was great.” It was like, “Well, how did ATT help you?” “Oh, ATT’s no big deal.” Then it’s like, “Well, actually, wait, wait back up there. Why is e-commerce on search doing so well, if ATT has no impact?”
If you wanted to steel man that you’d say “First of all, it’s based on keyword targeting.” They don’t pull that much information out of that. We don’t really know how much they personalize the search results on the basis of user history, we just don’t know, but it is keyword based, there’s context there, that’s what drives the relevancy scoring. Okay, fair enough. You’d say, “Well, there already is a policy that covers that, it’s ITP, that’s based in the browser. So, why do they need to have ATT apply to that?” Okay, fair enough. But there’s a lot of ways in which search is curiously exempted from this.
I mean, just to be super clear, when a search ad converts, Apple gets a share of that money.
ES: Exactly. Well, that’s not a steel man explanation.
Yeah, that’s the tin man explanation.
Apple’s Ad Network
Well, who knows? With this Apple perspective, the assumption is I think that this year, there’s a good chance Apple will kill the IP address tracking for apps, they’ve already done that in the browser. Apple is changing up their advertising team. Actually, I think the guy that used to run iAd back in the day is back in charge.
ES: That’s right.
One of the interesting things here is Apple obviously has their own ad network and this was a big deal in the Epic case — they believe they have access to all conversion data in all apps, because after all, it’s running through in-app purchase, which they legislate that you have to use. They’re using our purchase API, we get to know all the purchase data and given that, they have a perfect view into what conversions happen and what don’t. If they build out an ad network for apps, aren’t they going to make the same argument, “Hey, the app is running on an iOS device. We are serving the ad, it’s our data,” and then they’re going to have perfect visibility into who saw the ad and who converted and they can basically build an advertising product that to your point is so much better than what anyone else can do that if particularly when it comes to games? I know this affects e-commerce too, but particularly when it comes to games, why would you use anything else?
ES: That’s a head scratcher. Just their ability to claim that, that satisfies that distinction, our ownership of the OS and our ownership of iTunes satisfies that first-party custody requirement and so everyone else, sorry, go make an OS or go make a phone.
So many people that supported Apple in this are fundamentally opposed to the collection of data, which fair enough, I disagree on how damaging it is, but fair enough, I can respect that position. It’s frustrating the extent to which these people support this policy when actually the actual policy doesn’t stop the collection of data at all, it emphasizes the benefits of collecting data. It drives towards consolidation because the more touch points you have, the more places you have to collect data, and the more places to recycle that and show it, the better. This is a policy that, to your point, it doesn’t just hurt small and medium-sized businesses that are advertisers, it hurts small and medium-sized businesses of all kinds. You have to be big and large and have inventory to show and collect more data.
The other thing is if you want to actually do this modeling that we talked about in the context of Facebook, it’s not just that you need to know the purchase. Now, if you can’t know the purchase, the only way to work against that is to collect more data. So the outcome of Apple’s policy that all these people who oppose data collection were cheering is the collection of more data and the strong becoming stronger. Sorry, I’m interviewing you and I’m ranting. (laughing)
ES: No, no, we’re mutually converted, but that’s exactly it. First of all, the way that they think about first-party, third-party, that distinction is totally arbitrary, especially when they apply the first-party label to their own collection of this data.
Right. Part of the whole Facebook ecosystem was basically deputizing Facebook to collect data for everyone because no one wants that data, “I don’t want data. Please, Facebook, take care of it for me.”
ES: Exactly, operate the pipes for me on my behalf. The advertisers gave it up enthusiastically because it generated revenue for them. Just to hover on that point for a second, the way that they alleviate this burden for themselves, it’s not in a way that the consumer would ever assume is happening. The thing is like, “Oh, we show the ATT prompt, give the consumer a choice. That’s all we’re doing. We’re giving the consumer a choice.” Yes, fine. But the consumer would not assume that you are doing this on the back end because you have this very legalese style exemption for yourself. So, you’re giving the consumer the choice for every party, except for your own advertising network, which collects the exact same data! You haven’t changed the flow of data whatsoever, the exact same data is being collected and used for advertising targeting, but only by Apple and not by anybody else. That’s why I talked about this piece when I first introduced this idea of content fortresses, that’s the unintended consequence of this. It’s just like, “Well, then everyone’s got to consolidate” because we’ve got to put all the data in one place and we’ve got to bring all the content interactions into one place, such that the data that is emitted, those artifacts are first-party to somebody. The data is first-party to somebody and that’s probably the person that operates the ad network and if that’s the case, they can use that data to continue to target ads on my behalf.
I’ve got to sacrifice the independence of my product and be subsumed into some much bigger entity and that’s exactly what happened. The second ATT was launched, Snap introduced the retail product, Facebook already had Facebook Shops, TikTok introduced the retail product, Twitter introduced the retail product, they’re all trying to bring e-commerce onto the platform. Shopify said, their social sales were up 4x, from a small baseline, but that’s exactly what happened is that these platforms said, “Look, you’re dead without us. Why don’t you just take your product and operate it within our platform? Because at least then, when transactions happen, we can see them, we can oversee those and then we can use those to target ads on your behalf.”
Well, this is an obvious segue to Shopify. There are a couple angles here. One is the Shopify Audience’s product. I proposed a couple of months ago that Shopify should basically build an advertising network to service their merchants, but instead of selling ads, this network would be focused on buying them because Shopify knows all the purchases on their platform, and you actually pointed out to me, I think at the time, that they had actually hinted at this, this Audience product. The product has now been announced as of last week and it’s pretty interesting. What I thought was really interesting in your write-up about this was how you analogized it to Apple’s carve out for themselves. Walk me through that, because I thought that was very, very interesting.
ES: I think if you look at, and again you steel man what Shopify would say about this as to why it’s compliant? Well, you say, “Look, what Apple does is compliant,” right? They operate the platform, which processes the transactions.
Right, they claim that all transactions on iOS, no matter what app they happen in, Apple claims it’s their data.
ES: Right. I do want to be fair to Apple, they do have an opt-in for that personalization mechanism for their own ad network, but it doesn’t look like the ATT prompt. It says, “Do you want ads to be personalized to you?” Now, they gave a webinar last week where they talked about actually the opt-in rate doesn’t really impact the performance of the Apple search ads. I would say that well, first of all, if you look at the number of open positions at Apple on this ads team, they’re growing the team, they’re building it. Probably this personalization product is immature, but the other piece of that is because it’s all driven by search.
This is all about a future display network.
It’s now like Apple News and stuff like that. It’s not very big yet, but the obvious, long term outcome is ads in apps.
ES: Exactly, but I do want to be fair to Apple. They do have the opt-in prompt, although, they just worded it —
Which is very friendly. It’s like, “Hey, you can get better ads, da, da, da, da.” Whereas the tracking one is you’re going to be tracked across the web and go look at that commercial we sent about people talking about your divorce.
ES: Yeah, come on. It’s intellectually dishonest. So basically if you look at the way Apple does that, the way they prosecute that is they say, “Look, all these transactions are serviced by us and so therefore that’s first-party data to us and we can use it to target ads to people.”
Now, Shopify could say the exact same thing. “We’re the platform. These retailers operate on our platform. When a transaction happens, we process that transaction. We service the transaction. That’s first-party data for us.” Now what they do then — and your Article is very prescient and it describes it very well actually with the approach they take, even though it was published months before they released a product — is they package that data up and they look for lookalikes. If I’m a shoe retailer, they look at the people that have bought shoes and they package that into an audience. They hash the identities, they hash whatever the identifiable information is, they upload it into Facebook for that retailer’s advertising account. Now, when that audience is uploaded, it’s not downloadable by the retailer. The retailer never gets custody of it, so there never is that first-party, third-party —
This is the clever part that I missed in my Article, because I presumed that given the ban on your sharing lookalike audiences and targeting with third parties that Shopify would actually have to go in and buy ads, which they’d have to build an interface for retailers to buy ads on Shopify and Shopify would buy on Facebook, which obviously is a total kludge, you don’t want to do that. The thing that I didn’t think through which makes total sense is, “Hey, no, we’ll give you all the data you need to buy the ads and oh, by the way, we’re going to package it in such a way that you can’t download it. You can’t access it. It follows the letter of the law, but you get the same outcome without having to us to have to build a weird kludge in the middle.”
ES: Exactly. Just to zoom out a little bit, I think that letter of the law qualifier is important here, because my sense is that these big platforms have become very frustrated with the reticence on the part of Apple to clarify anything and to work with them. My sense is the attitude has changed from, “We’re going to honor the spirit of this law and we’re going to do it in a fulsome way to show that we’re good actors.” And I think that has transitioned to, “No, we’re going to follow the letter of the law.” The letter of law, by the way, is super ambiguous because that’s what Apple does, Apple likes to operate in the dark, they don’t like to have a spotlight on their internal policies around this stuff. They like to be able to make decisions on a case-by-case basis. I made the point in my article, ATT is not law, ATT is a platform policy that is litigated in the dark on a case by case basis and Apple likes having that optionality. I think these platforms are doing this to say, “Okay, Apple, we’ll do it your way. We’ll honor the letter of the law here because the letter of the law is very vague and it’s not very instructive.” So now, you have to tell us where we’re violating and by the way, we’re Shopify. We operate on the web, what are you going to do? What is your punishment mechanism?
And then you have the conversions going in the background, which is all server-to-server anyway. I want to double down on this point because I was just feeling a little bad ten minutes ago. I’m ranting too much. It’s too anti-Apple. People are going to be upset, but I think actually that’s representative of how the whole industry is feeling right now, where I think that Apple got a lot of the benefit of the doubt about a lot of this stuff, but it does feel like a line’s been crossed where almost everyone that you talk to — and this isn’t Facebook people — this is people all over the place — just feel it’s so unfair and it feels so self-preferential.
Even if you want to give Apple the benefit of the doubt, especially when you see that everyone who was saying at the time, “Oh, Apple’s going to build their own ad network” and people were like, “Oh, no, they wouldn’t do that.” The fact they’re so actively investing in it, I think that, yeah, we probably overdid it a little bit, but I don’t think we’re alone. I feel like that’s representative of the way a lot of people are feeling. I hear from a lot of hardcore Apple fans on some of this stuff, particularly when I wrote about the way they collect data and the first-party, third-party distinction, and they’re like, “I didn’t realize this was happening. I thought this was about not collecting data.” I think a lot of people have woken up to this recently.
ES: I explain this to people and people will be like, “Come on, you’re exaggerating. That’s got to be hyperbole.” Well, the thing is, though, if you wanted to find a vector of commercial competitive attack on Facebook, there’s none better than this very murky, esoteric, ad tech arena. Very few people understand it, the commercial impact’s going to be gigantic, but it’s going to unfold over time. You can point to a lot of other mitigating factors. “Well, no, it wasn’t ATT. It was the broader tech rewrite”, which again, that’s true. You point to any of these stocks and say, “Well, yeah, part of that’s ATT.” How much? I don’t know, because the broader ecosystem is melting down.
I made this point on Twitter a while back and I wrote something like there’s probably 1,000 people in the world that have agency over these decisions at these big tech companies that will be responsible ultimately for the future of the Internet. A lot of people corrected me, it’s like, “Now, it’s probably closer to 100”. So few people understand, and my sense is those 100 people probably, if you think about that on a go forward basis, control more economic capacity than OPEC does. This is a massive, massive part of the economy and it’s being controlled by 100 people because so few people work in the space, have worked in it, understand it. These are really important decisions now that any regulation is far behind on this, and the thing is the platforms can lead with these platform policies are not laws, but that impact the economy in very momentous ways.
I think to your point, and to defend Apple, all this murkiness and fuzziness and happening in the dark favored the ad ecosystem previously. I think you’ve been very consistent on this point, that there was a lot of bad stuff happening, and I think one of the frustrations is a lot of that bad stuff was mostly in the ad broker ecosystem and it wasn’t Facebook. Somehow Facebook got tarred with the bad actions of a whole bunch of other folks, which is another thing actually that I think has been coming out more and more the last little bit. There was an article the other day about Grindr and how all the profiles are being sold and stuff like that. That’s not a Facebook issue, that’s actually Facebook fixing the issue by keeping everything on the platform, but the broader point is the ad ecosystem had this coming. We’ve gone from one extreme to the other extreme and so there was something there that Apple was justifiably attacking.
ES: Yes, I wrote an article and I said, “Look, IDFA is the hydrocarbon of the mobile ecosystem. It’s got to go. It’s being abused. It’s being used for purposes that are abusive or hostile to users. It’s got to go, but not this way.”
We never got to the rest of the Google piece. What they announced is, “We’re wholesale getting rid of the GAID 100%. We’re not going to have an opt-in, it’s gone. We’re going to give you two years, but we’re going to replace that attribution functionality that you use the GAID for with these other tools that will be functional, but we’ll also preserve privacy and we’re going to do it out in the open.” It’s going to be collaborative process. That’s the Privacy Sandbox, they did it on for Chrome, now they’re doing it for Android, it’s going to be out in the open. “We’re going to collaborate with the industry. We’re going to come up with solutions that work for everybody, but that are also privacy protective.”
I think, for the most part, they’ve been good actors there. They had FLoC, which got pushed back on, and now they’ve got the Topics API. They adapted, but that’s the way you do this, it can’t be a War in Iraq style thing, where we’re just going all in and we’re not going to build a coalition. We’re not going to bring on partners, we’re not going to take the temperature on how people feel about this, we’re just going to do it. You can’t do that. Again, this economy’s way too big and then you also can’t do it in the way that it’s just blatantly self-preferencing.
Regulation and the Future
I had a list of questions, we’re stuck on this one, but it’s a good discussion. How does this play out in the very long run? Something like the Topics API, is that going to work for people? My theory — relatively unfounded but just a broad principle — is that Facebook and Google are probably still better placed just because they have the most resources to figure this out. They’re going to throw machine learning at this problem. That’s a very expensive proposition by the way and obviously pending Facebook’s ability to keep users, but assuming they do, or is there going to be a way for other folks to break in? Honestly, it feels like the smartest possible thing to happen here is Facebook should just buy Shopify. I don’t think there actually is a competitive concern, they should take it to court, is there any other response than just massive consolidation?
ES: Ask Lina Khan, I don’t know that.
Honestly, I would hate it, I hate it just as an observer. I love companies like Shopify that enable the small and medium-sized businesses. But if I’m Facebook, I do it and I take it to court and I don’t see how they lose. It’s going to be expensive, it’s going to take years, but at this point, I don’t know, that’s neither here or there, let’s assume that doesn’t happen. Leave that one to the side.
ES: I think if Facebook could accomplish M&A, they should do that. It’s funny because I was thinking about this today — right after the WhatsApp acquisition closed, I was at this conference and the head of M&A at Facebook, the guy that did the WhatsApp deal was there. Everyone was talking about him behind his back like, “Can you believe they paid that much for that? What were they thinking? That’s the dumbest acquisition ever. They paid that much for WhatsApp?” Everyone pilloried that acquisition, everyone thought it was a bad idea that they paid way too much for it. Now, all of a sudden, in retrospect, it was a great idea. But at the time, no one saw that as this brilliant peering around the corner thing. Everyone thought the price is excessive, there was no strategic value necessarily, there was no synergy there, there was no way to probably monetize it, everyone thought that that was a bad acquisition. Well, in hindsight, it was a great acquisition, but it’s just weird, I think, to then apply the standard of Facebook that says you grew through acquisitions that everyone agreed at the time were stupid. Clearly you have so much great insight in or excellent executors that whatever you do is probably a good idea so we’re going to stop you from doing anything, because that’ll just make you bigger and wealthier and more successful. That just seems like a weird way to regulate business.
Well, there you go, what are you going to do? Is Facebook going to figure this out? What’s going to happen there? I guess to the regulatory point, you wrote about the Digital Services Act(DSA) and its impact. What role is regulation going to have on this? The other potential regulatory issue is on the other side, where Apple actually does get taken to task for this or this is all going in one direction.
ES: I don’t think that there’s any regulatory scrutiny of ATT, I think that ship has sailed. There was some in the UK, the UK CMA, they had a report and they did this app market competitiveness report. This was three or four months ago and they had a whole subsection on ATT and they basically said, “Look, this is a blatantly self-preferencing move.” A lot of their problem was the text of the prompt and the dark patterns of having the ask app not to track. It was higher than the other option, but I don’t think that’s going anywhere, my sense is nothing’s going to happen with that.
There’s some great stuff in the DSA, what was most meaningful to me about the DSA was what was not in there. Right now, there was just a money gun pointed at Brussels by the big tech companies in the lead up to DSA as they were negotiating the text, and so a lot of that stuff got watered down. But my understanding was that the original proposal is they wanted to ban targeting advertising wholesale, just ban it, “You may not target ads to people”. That is closer to the text of the Banning Surveillance Advertising Act that was proposed in the US, but that got watered down and effectively, it got replaced with “You can’t actively target ads to children”, which I think any reasonable person agrees with. I don’t want ads targeted to kids, that’s an obvious change to make.
There’s two pieces to this: How does it get interpreted, and then how is it getting enforced? The problem with the DSA is you’re going to need an army of data scientists to do all the stuff that they want to do with just having visibility and transparency into the data and the algorithms. How do you do that without 100 data scientists on staff? I don’t know how they’re going to hire those people.
But the other thing is, how do you interpret that? I saw the final-ish text of the DSA, and before that version, there was a concern that the way that the prohibition on targeting to children was worded was that it could be interpreted as “Unless you know for certain that this user is not a child, you may not target ads to them”, the way now you’d interpret is that “You may not target this person on the basis of knowing that they are a child — knowing their age and knowing that they’re younger than 18”. The other issue in Europe is there’s a whole bunch of different interpretations of the word “child”. The age of consent, for instance, across Europe is wildly different from country to country. But now it seems that the interpretation of this is that “If you know for certain that this user is a child, you may not target ads to them on the basis of their age”. I think any reasonable person would agree that’s a good idea, we should not do that.
But I think to answer your question about Facebook, yeah, they’re better positioned, Google’s better positioned, they’ve got the data science teams. My understanding is when this happened, Mark Zuckerberg by fiat threw 2,500 engineers at this problem. The ability to do that is unmatched, who else could do that? The other thing is if you look at their OpEx, yeah, a lot of the OpEx spend is going to Reality Labs, but a lot of it is to go into the ad platform and beefing it up, they’re spending a tremendous amount of money. My sense is there’s a two-year timeline. Right now, there’s two parallel tracks, one is fix the measurement, fix the rubber dinghy, fix the hole, because otherwise, you’re going to drown. And that’s what they’re doing.
I think they get it to 100% maybe in a year or whatever, but they get that back, they claw all that back. The targeting piece, it’s a much heavier lift, but they’ve got the people, they’ve proposed some really amazing things. One of their teams proposed this thing called IPA, interoperable private attribution. Basically, there’s almost a clean room, but it’s like a clean room co-op. So you’ve got this server that sits in between all these platforms, and they upload their data to the server all in an encrypted way, they apply differential privacy, it’s very privacy safe, then the advertisers send in their conversion data. It gets matched in this environment where nobody has total insight into the two pieces of data and then the results get sent back to the ad platform, it’s a very elegant solution. Now, I have no idea if Apple agrees that that’s possible or that is allowed by ATT. I have no idea if they’re able to get a broad collaborative effort across these ad platforms, but if they can achieve what they propose, it’s a beautiful solution. It’s very elegant, it’s going to take many years to implement.
The Ad Agency Opportunity
There was a tweet about retailers like Macy’s Kroger and Michael’s having their ad businesses. You quote tweeted that and said, “Everything is an ad network.” This reminds me of a situation a century ago where there was an increasing number of places to advertise geographically, and this gave rise to ad agencies where there was a one place for advertising to interface with, and they could manage all this other stuff. Is there a similar opportunity today for a new kind of digital ad agency that’s not the same ones that came before, but it’s almost like what Shopify is proposing to do for their own merchants, or is this really something like an AppLovin is doing or Unity is doing? You have to have your fingers in as many apps as possible and in as many sites as possible, you have to declare this is all first-party data and then you’re going to figure it out.
ES: Yeah, I think so. You just go back to these big platforms disintermediated the ad buying process, and now it gets intermediated again, because the advertisers don’t have the capacity to do that stuff, to just manage campaigns across a lot of different networks, maybe that’s what happens. The thing with the retail media network phenomenon is Facebook and these big platforms, they were the everything store for ads. I could reach people with an ad for any product and just the vastness of these platforms and their ability to target relevant customers made that just the perfect place to go to advertise anything and now that supremacy has been deteriorated. So Costco has an ad network, if I want to advertise Jumbo, a box of wet wipes, tater tots, or Hot Pockets, I’m probably going to go to Costco’s ad network, because that’s a more relevant place to do it. Macy’s has one, Best Buy has one, Home Depot has one. So, if I want to advertise my lumber product, I’m probably going to go to the Home Depot ad network versus Facebook today, because they’re not going to be able to find those lumber buyers as easily as Home Depot can in a contextual way. And the list goes on, Kroger, Target, Best Buy, Walmart, it makes sense, you subdivide the ad space into contextual relevancy for certain types of products and therefore certain types of advertisers.
That feels like the future of the next WPP is this opportunity. Facebook to your point, I wrote about this a couple years ago, I’m repeating you, if the only place you buy ads on is Facebook, then why do you need an ad agency? There’s a lot of places you to buy them, then there’s much more of an opportunity.
ES: On the ad platform side, I think this stuff is happening. The thing is the regulation is catching up with the platform policies, and so all that stuff will get unified at some point, and it’ll probably look like something that sits between the DSA and the Banning Surveillance Advertising Act, I think. I’ve been told by people that Banning Surveillance Advertising Act, it’s not even going to get considered, it’s not a legislative priority, but let’s say that gets watered down a little bit. That’s the law of land in the US, it’s probably stricter than the DSA in some ways and so that becomes the global law of the land, probably that happens. Then it doesn’t really matter what the platform policies are because this is going to be more strict.
The ad platforms will figure this out, but it’s probably a two to maybe five years on the outside proposition. Advertisers — what are they going to do in the intervening years? A lot of this spend is just gone. So, what you did as an advertiser prior to ATT was you just said, “Hey, Facebook, figure it out. I’m abstracting all targeting to you, all audience development to you. I have a product. Everybody enjoys it in the same way, everyone sees it in the same way. Facebook, go promote it to the most relevant people.” Well, Facebook can’t really do that anymore. So, what are you left with?
Now, you get what I call the traffic blob. It’s this totally heterogeneous group of people there, there’s not a whole lot of commonality. There’s definitely not as much relevancy as there was before. And so, you as the proprietor are forced to then parse out the people that are most interested, the people that have the most intent, and personalize the experience to them in a way that allows them to express their intent monetarily. That’s a difficult thing to do. It’s really, really difficult. It takes a lot of resources and those SMBs? Forget about it. That was the power of these systems, it’s like they allowed the SMBs to operate with this high fidelity firepower of a Facebook.
It let a small company compete on an equal ground with a big company and that’s gone now.
ES: Yeah, it’s just gone. And so now the advertisers have to build the measurement piece to say, “Okay, we don’t have the user level data anymore. We need a more probabilistic signal.” That’s tough, there’s things like media mix models, you can do incrementality measurement, lift measurement, that stuff takes data scientists. If you’re a three-person DTC company, forget about it, and then it’s the whole personalization process. “Well, this person is an unknown quantity to me. This person coming in, they might be super primed, they might be someone who wants to spend a lot of money, they might be someone who’s ultimately going to be economically worthless to me.” I’ve got to figure that out and that’s actually a difficult task and all of that is left at the feet of the advertisers and that’s a big problem to solve. Now, no one can just sit around and wait for two to whatever years for Facebook and Google and all these other companies to figure this stuff out.
Yeah, we’ll leave it there. To the point earlier, the reason why I get worked up about this is I’m so excited about the Internet as an opportunity for new kinds of businesses, particularly the small businesses where the whole world is your market. You can serve a niche, but you’ve got to be able to find that niche and the casual way that not just Apple killed it, but the way people cheered them on, it’s been frustrating. Facebook’s going to be fine, Google will be fine, yes they took this big hit, they’ll figure it out to your point, but this world in which you could just be a one- or two- or five-person business and actually compete, it’s gone. And that makes me sad.
ES: Yeah, same.
All right. Well, on that note, I’m already sad, because the Bucks just lost, but what are we going to do? Thank you for taking the time. We went long as usual. We only got through a quarter of my questions, but we’ll have to do this again soon.
ES: Cheers. Thank you for having me on. Appreciate talking to you.
This Update will be available as a podcast later today. To receive it in your podcast player, visit Stratechery.
The Stratechery 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 subscriber, and have a great day!