Interview. Eric Seufert on Whether AI Will Destroy Google's Search Monopoly

0:06 Is AI Chat Disrupting Google Search? Hello and welcome to The synopsis, a podcast for professional investors. I am really excited for our conversation today with Eric Suford, who is an independent analyst and creator of Mobile Dev Memo. 0:22 He writes a lot about the ad tech ecosystem. In fact, when I was at Capital Group, he was the expert we often called on to figure out more about what was going on in AD tech So, Eric, welcome. 0:35 Speaker 2 Yeah, good to see you. We were just talking about the, the, the, the time that you, you very graciously brought me out to Palo Alto to The Four Seasons. We had a big dinner there. That was a very nice event. And so it's been a while since we've seen each other, but good to see you. 0:52 Speaker 1 Yeah. And I'm excited to talk today and kind of the beginnings of this podcast was there was a one of the past podcasts we had, we talked a little bit about Google search and some of the, we'll say, challenges facing Google search and you kind of had your own thoughts on that. 1:10 So just to kind of set the stage a little bit before we get into it, right now, Google Search, as we all know, one of the greatest businesses of all time, very high margin business, very scalable and it is ubiquitous. 1:25 Something that has been changing potentially is the rise of AI Chat. And so there's a few different sort of things that could potentially change for Google Search alongside the rise of AI Chat 1 is just that now there are new competitors, new interfaces for people to potentially go to instead of traditional Google search, which could become disruptive later on. 1:49 But then second, there is also the form factor of search changing as it becomes more conversational and maybe the ability of advertising to be injected in changes to the negative for Google. And so I just kind of wanna set the stage right there and get kind of your thoughts on what you think about the changes happening to Google search are and where you think it could kind of go in the future? 2:14 Speaker 2 Well, I, you know, this is a, a favorite topic of mine. So, so I wrote a, a long, or I wrote a, a, a series called Google's Gambit, right? And I, I, I want to say the first entry in that, I mean, it would have been sometime in like late 2024 or 2025. 2:31 And and so the, that this was when you know, pessimism about Google with respect to like the kind of changing paradigm of like chat based, sort of like natural language based chatbot based interaction models for content discovery. 2:54 Was that like the fever pitch, right? I mean, like, this is Google was just seen as a victim of Chachi PT and it, it was going to be just overtaking completely. Search is going to be overtaken completely by this chatbot paradigm. And I said, look, here's, here's what seems to be happening. 3:09 Like Google is capable of navigating this. I mean, I think in general, and I mean, this is like painting with a very broad brush here, but Google is seen as this kind of like very staid, sclerotic organization, very slow moving, you know, not very flexible. 3:27 And, and, and, and maybe maybe that characterization is fair in some respects, but Google did seem to be adapting to the moment, right? And so you know, what they had announced was AI overviews where they were going to sort of take over the above the fold space and search with an LLM generated overview of all the content that you otherwise might have navigated to. 3:51 And I, and I, and I remember at the time I called this like what it was, it was in the very early stages of this and I don't even remember if they'd launched yet. Maybe they just announced it. But like I said, look, they have this ability to adapt to this moment and to execute this like total ship of Theseus transformation of the search experience. 4:09 Like there's no reason why they don't have the right to win that interaction format, right? That interaction model. Like there's no reason why they don't have the right to it. Now, if they can't execute well, well then they won't. But there's no reason why they don't have the right to win that. 4:25 And my sense and I, you know, track the progress of this as it's evolved. And my sense is that they did right. Now, that doesn't mean that ChatGPT can't become popular. It is, it's a, it's a billion Wow, right. So it's a very popular product. That doesn't mean that other chat bots can't become popular and, and, and see a lot of usage. 4:42 But but my, my, my instinct at the time, and I think this is proven out, was that Google could adapt to that moment and they could, they could transform the search experience into what I called a, a, a distribution mechanism to a content sync, right? 4:58 So to take all of that engagement that they used to, they used to funnel out of the search experience, which is the whole point of search. The search did a good job if it got you off of the page with us without having to click multiple times or refine the query or go through to the second page. 5:14 It like search did its job really well if you click that first link and the company did really well if that first link was the most relevant possible destination and also an ad, right? Everyone won if that was the case, including the advertiser. We can get into that in a second because I think that's important when you think about Agenta commerce. 5:32 But every like everyone won if that first link was the most relevant in an ad, right? And so the whole, the whole purpose, the whole, the whole objective for search was that to get you out as quickly as possible. And I said, look, they can transform this into this content sync where actually the whole purpose is to keep you there, right? 5:50 And if they keep you there and they get you refining the queries and they get you tuning your queries and they get you asking follow up questions, then they win because they keep the engagement and they get multiple opportunities to show you an ad. And I think that's what's happened right now. We see the engagement for AIO reviews AI mode is very impressive. 6:07 It's 2.5 billion Mau for AI overviews. It's 1 billion Mau for AI mode, right? They're capturing all that engagement, all that engagement that used to leak out. They capture it, right? And they had, they show ads now they show ads in AI overviews. They show ads in a mode. They just introduced Universal Cart, which I do think is a better model for agent to commerce than what Muse is talking about. 6:25 But maybe that's for later. But they've done all these really innovative things that I mean would be inconsistent with this characterization of this very sclerotic, slow moving, inflexible companies. I think they've done a really impressive job of transforming essentially the revenue engine of the company into something wholly different than it was three years ago. 6:45 Speaker 1 So I like that kind of overview to kind of move a little bit more into thinking about some of the potential risks to Google during this transition. Because right now it seems like they're in this position where they're still benefiting from the legacy search business they have. 7:03 And a is potentially been boosting that too, with their ability to, to better target ads and more contextual advertising. And you could talk a lot more about that more intelligently than I could. But I we're also now kind of at the beginnings of consumer behavior potentially shifting. 7:20 And it's hard for me to think about a world where there is a ChatGPT that's not going away. There is Claude, there is Grok. There are all of these different modes for people to move their search activity too. And I, you know, listening on their last earnings call, they talk about how search queries are at their all time high, but they never specify if they're including in that the AI search queries. 7:45 And within that you can find, you know, it's third party data. So you have to take it with a grain of salt. That or you know, there is one data point from Eddie Q's from 2025 from the Google antitrust trial where he was testifying and he notes that Safari search searches were down on Google and that Google said, oh, that's not really true, but that was his testimony that that said that. 8:05 So there are some points that kind of show that potentially traditional searches eroding. And I think you kind of will see that maybe is that being a possibility? Except you're trying to make the point that this future of AI with AI search is going to be something, as you call it, a content sync that's going to be able to retain users longer and have more opportunity for advertising. 8:29 Have I kind of conveyed that accurately? 8:33 Speaker 2 Well, I mean, yeah, look, I mean, you can, you can interpret that EQ comment in a number of ways. Look, I mean they did $63 billion last quarter, 17% year over year growth. I mean, if that's what erosion looks like, then, you know, I want to reincarnate as a as a, as a Cliff side in California because erode me. 8:54 But I think like the the question is like, what is search and does it matter? Does it matter what you call it? Right? Is that just semantics? Because if AI overviews has essentially totally display search and I think, you know, there's other data that supports that it has. I mean, the revenue is growing. 9:09 They had, you know, they said all time high, I mean, during the World Cup, right. So July, you know, 2026 is world time is an all time high. OK, fine. Maybe that's a special, you know situation and we don't know what it looks like that, but fine to to have all time high searches in the last quarter. 9:24 I mean to what extent can you actually say that the business is imperiled? And the question is like for any interaction model that you think could ultimately displace Google, forget about whatever you call search. Let's lump AI mode, AI overviews and the traditional search into just this bucket that we call search because I think it's all kind of monetized that way anyway. 9:46 But what could displace Google as the dominant player here? Well, is that just chat bots, standalone chat bots? Because they've got one, they've got Gemini, right? You know, so like, is that a, is that an agent? Maybe like they're, they're maybe like the, the muse model displaces the need, you know, or obviates the need to go to like a search engine or to just search anywhere. 10:10 But, you know, the sort of the central agent that you use. Well, fine, but it's not like Muse doesn't have competitors and there's no reason that Google can't also release an agent. And so my sense is like, it just, it just strikes me as as unsubstantiated to think. 10:28 And in like, in like might, you know, maybe. And again, like this was two years ago when I started the Google's Gambit series. I the pessimism was, you know, was, was overwhelming. I mean, like there was like, there's this, you know, tremendous chorus of, of, of, of bear sensitivity around Google because they seem vulnerable. 10:48 But like, we're not that early, right? Cha Cha PD is 4 years old, right? And you know, it's far and away the largest chat bot in the West. And we still get $63 billion a quarter at 17% year over year growth, right? So like, to the extent that it's eroded, I don't know that the, the erosion has happened as a result of these external chat bots. 11:07 To the extent that it's eroded, it's probably eroded because of Google's own reinvention of search as this new thing that is clearly monetizing very well. I mean, they've said it monetizes at parity and they said last quarter that search queries are at an all time high. Now how we define those things might, might be different if you if you classified AI overusing AI mode as this distinctly unique thing. 11:29 But at the end of the day, they call it search, I'll call it search. 11:32 Speaker 1 Yeah. So I think the distinction between search and AI chat is because there could be a difference in how these interfaces end up monetizing and what feels most appropriate to a user in these circumstances where when you are on Google search, it's totally appropriate to have these sponsored ads pop up and you could go ahead and click on them. 11:54 Whereas if you are in a chat interface, it feels like the there's certainly opportunity to show ads, but it does feel like you might have to be a little bit more selective with that, especially if you are trying to continue to garner users trust. And I think the idea of having ad run chat, which, you know, it exists today in open AI, but it does, you know, worry us a little bit into the fact to the extent that open AI's ads are really bad right now, It's actually kind of a positive thing, at least from the user trust perspective. 12:23 Because if I'm in an ad, if I'm, you know, on ChatGPT and I'm asking chat about, you know, nutrition or health or something, I don't want to believe that, you know, the recommendations or the things coming up are somehow going to be influenced by their ad dollars. And of course they're going to say we'll never do that. 12:41 We're, we're not going to have, you know, our model waiting for, you know, the advertising aspect. Instead, you know, the answer is going to be agnostic of that. And then below that, while the separate ad system that maybe has is picking up a little bit on the context of the conversation. Nevertheless, I, I do think that that gets to an area where people are a little bit more uncomfortable. 13:00 And you know, correct me if I'm wrong, but right now Gemini doesn't show ads in their conversations. And so if we are looking at, you know, Google's search, search and other as they call it in that category. So it does include some other things. If that is, you know, at an all time high, it's not based off of AI conversations. 13:19 And that threat to me, even though maybe we've been using AI regularly for a long time. We're kind of at the beginning of that S curve and that penetration is still very early. You know, my dad just learned about open AI, about ChatGPT a few months ago and just started, you know, using that daily. 13:35 And I just feel like this disruption, even though we're maybe very used to it, it, there's still potential for it to be pretty early. And I'm not convinced that if you were going to see erosion in Google's financials, I don't believe it would have happened so far. I think the the first impact is that AI makes search better before maybe consumer behavior down the line changes where people start these searches. 13:57 Speaker 2 Early every early four years, a billion weekly active users, 2.5 billion monthly active users on AI overviews. Are we early? I feel like we're right in the middle of the S curve. Like this is not a billion, a billion one out of 812%. We're at a billion. 14:12 This is not that early. How many people could potentially use it at all? I mean, we're, I think we're right in the middle of the S curve. The, the early was two years ago. The early was when people were saying search is dead because Chechi PT is taking off. I don't think we're early. Look, the trust thing I think is a canard. I don't believe AI don't believe that there is the I don't believe that there's a vulnerability to from the user side toward the toward the usage of ads with respect to like deteriorating trust or the sense of objectivity. 14:47 I just don't think that that's the case. That's a design issue. Now, people could fumble the ball and they could insert ads in a way that are that are that that that test the boundaries of that trust because they're not clearly marked. But my sense is like, if you look at the way ChatGPT is doing it, ChatGPT is a billion dollar, you know, annualized revenue with, with ads. 15:09 I mean, like it's not even early with that. I mean, they're they this is 200 days in and they're a billion dollars, right? I mean, I think like the timelines accelerate when you talk about AI, because a lot of the AI is being, with the exception of chat GPTA lot of the AI enablements being tacked on to already scaled products, right? 15:26 Like Meta AI exists in all those platforms, like Gemini exists in this entire, in the almost the entirety of its product suite. And so my sense is like they're so first of all, it is a misconception that consumers don't like ads. They don't like bad ads. 15:41 But there's there's enormous pool of empirical proof that and and and research that consumers a benefit from ads and perceptibly benefit from ads. They know that they benefit from ads. And so there there's a misconception that consumers don't like ads. 15:56 BI mean I think like this idea that having ads in a chat bot necessarily by definition erodes the consumers trust or is a hostile act by in and of itself to the consumer I think is a canard. I just don't think it has to be that way. I don't think it is that way. 16:12 I think chat ChatGPT has proved that that's not the case. And so, you know, I don't know that like we are at this like, yeah, but stage of the implementation because yeah, yeah. But as soon as they implement monetization, the whole business model is going to fall apart because users will will vote with their feet and abandoned the product. 16:34 Well, if that was the case, we would have seen that with ChatGPT. And then there are ads in AI overviews. There are ads in AI mode, right? And so we know what that looks like. Now. The other thing is what just became available to US users last week was universal cart, right? That's a form of monetization. They've got direct deals there, that's ads, right? 16:49 And what that does is that, you know, has this what just like what it's called, it's just what it sounds like. It's a universal cart. It can monitor products that users care about. It can consolidate the cart across the retailers. When the user's ready to check out, then they can do that and then can either hand off to the retailer directly if the retailer wants that, or they can check out right in the Google interface. 17:10 I mean, that's, I mean, that's not necessarily an AI application, but that's a way to sort of bring more and more of the engagement into that, into that interface, that, that interface that used to drive the engagement away. So like, my sense is like, you know, the, the trust thing, I, I feel like that was, that was an open question. 17:29 It like in the instant checkout era of like open AI really, really, really trying hard to resist integrating with ads. But like once they did it and once they saw the tremendous amount of success that they did and and they executed at the pace that they did, which is a clear signal to me that they believe this is the future of their product. 17:46 Like my sense is that you could kind of at the very least say this might be a concern if it's executed poorly, but when it's executed well, it's not a concern at all or the concern is overcome by the opportunity that is presented. And my sense is like, there's also you could have, I also just don't know that you couldn't have said that with search. 18:07 I don't. And then first of all, the thing with like the whole thing, yeah, sure, if you're searching for like sensitive topics, like there should be no ads there, but there are no ads there right now. And there are no ads there in other interfaces, right? Like so in like other other, other other products when, when ads are present, right? Like there are, there are platform rules against this. 18:26 There are laws against this. You can't monetize sensitive topics with ads And so fine. Like so, so that you know, if, if you know, if you're searching for like, you know, I don't know, like you go down any number of examples, but like you saw an ad, yeah, that would that would bother you. That would that, that would, that would be disconcerting. 18:41 But like, that's not going to be the case because it can't be the case. And it and it could be the case in in other products too, but it's not the case. And so we could just borrow from those examples. The thing is like, if you look at what what Google historically has said, it's like 20% of our searches are even monetizable, right? 18:56 Well, then, then maybe that's the surface area for using the context of the conversation for ads. But now let's say that you know, the chatbot experience like, well, I've got 20% of the chatbot conversations that are monetizable. But because I'm building a modern ad platform with a Cappy with the Pixel, these things that didn't really exist when search got started, the Pixel did, but not the Cappy, then I can actually just do display based on consumer behaviors and off platform behaviors. 19:22 And so like I get this opportunity that Google never really had with search to monetize more of the surface area of the product interaction. So my sense is like, this is an even Better Business. It doesn't necessarily need to be contextual. And so when there are these conflicts, you can just sort of you can just sort of punt and maybe you don't serve an ad right against that, right against that conversation because again, that would be disconcerting. 19:44 But you don't have to own like if the person only ever searched for like health issues and maybe you could never monetize them, but they could search for health issues in the main. But whenever they don't, if they search for something else that's not monetizable directly from the context, you could surface an ad based on their off platform behaviors, right? 20:00 So, so my sense is like the trust thing is an execution problem. And if you avoid doing it poorly, I don't think that it is a fundamental. Hurdle for having success with ads, and I don't think it's something that necessarily will bleed engagement to other platforms that don't implement ads. 20:17 Speaker 1 Yeah. So just kind of responding to all of that, that was a pretty good that. 20:21 Speaker 2 Was long. That was long. 20:23 Speaker 1 No, that was good. So first thing I'm with you that I don't think that advertising is itself like some sort of evil. I think that it generally can be additive to the product experience. I think that, you know, metas, Instagram is a very clear example of that, that people actually enjoy those ads when they're well executed. 20:43 I, I don't know whether or not AI is a different thing, especially as we get more into agentic AI and the potential to have an AI agent run more of your life. The kind of idea that that could be ad supported, it would make me a little bit uncomfortable. 20:59 I would rather just pay for the subscription, in which case it's a different kind of competitive dynamic whether or not it ends up being Googled Gemini that that wins that or different AI agent. But I hear you. They still have, you know, a place to continue to, to win That kind of broadening out to the the bigger picture here. 21:15 Can Google Rebuild One of the Best Businesses Ever? And the bigger threat to Google is because right now we're kind of talking about how Google has this very good position in AI overview in Gemini and they have this good ability to deal with AI as the world goes more to AI chat. 21:32 However, as evidence for this business model kind of working, we're still relying on the financials of Google search, which is still where, you know, the vast majority of all of this Google search and other revenues coming from because AI isn't really monetized yet, barring, you know, some some subscriptions for Gemini. 21:51 And so my question still is, are they really going to be able to navigate going from one of the world's best business models ever to now, you know, a chat interface, which we know is more compute intensive, which, you know, we talked about. Maybe they do have some of these issues figuring out what the appropriate time to show ads are. 22:11 And then there is also, you know, the fact that maybe they do miss out on a good number of search queries where, you know, companies just are, are bidding on their own name. And you know, I, I don't know exactly what number that is of searches. I think there was some research paper that that showed it was somewhere around anywhere from like 25 to 1/3 of searches could be just companies bidding on their own names or our competitors names defensively. 22:36 And so that is still a portion of that. And so my thinking isn't that, you know, Google is going to be a 0 or anything by that by any means, but just are we really going to go from one of the world's best business models to a business model that just happens to be as good or is really even better? 22:51 That's that's to me the question mark I have. 22:56 Speaker 2 Well, OK, so, so you're talking about like branded search, right? And I, I, I haven't seen any research on this either. I don't, I don't know that it exists, right? I mean, outside of the walls of, you know, Google, but empirically from anecdotally, from my experience, branded search is a tiny minority of, of ad spend budgets because it just tends to not work. 23:17 And if you see it being done, it means it does work. I mean, companies are very good at doing incrementality testing, incrementality testing in digital ads. Is, is, is when I sort of test the impact of my spend against some or sort of let's say I engineer a counterfactual for my spend, right? 23:40 So I engineer this opportunity to observe what would have happened if I didn't spend the money. And I see if, OK, well, if I didn't spend the money and I still, you know, and the impact was the same, or like I still saw the, the benefit that I saw in the instance where I did spend the money, Well, then there's no incremental benefit from spending the money and I shouldn't spend the money. 23:58 And you know, I think there's this like there's, there's there's this kind of pervasive belief that most marketers are like really bad at what they do and they're really gullible and they're really wasteful with their ad budgets. And they just sort of spend in a way that is kind of reckless and has little impact on the revenue. 24:20 And a lot of the ad platforms are the beneficiaries of that because they're not actually delivering any revenue. And I think that's just demonstrably not true. Now, I'm not trying to set up a straw man. I do believe that. I'm not saying that's what you're saying, but I'm saying there is a pervasive belief there. And like, what I would tell you is that like most advertisers that are spending beyond some kind of like very minimal amount of money are pretty good at incrementality testing because it's not that hard to do. 24:48 And the ones that aren't that good at it are, are spending money in ways that are sort of like in and of themselves easy to track against the revenue contribution. And so there's kind of not that much room to just be totally wasteful. 25:04 Now, I like there, there, there are people that will make the case that like all digital advertising is fraud and it's essentially all fraudulent. And like these platforms deliver 0 value. I mean, those arguments are easy to to ignore on their face. 25:15 Speaker 1 Sure, sure. Just on the incrementality testing that when you're setting up that test you would set it up as running a sponsored ad for your brand name versus not, right. And the counterfactual in that case isn't what if the organic search result was, you know, there were no sponsored ads upfront and organic search result was just there first. 25:34 And so then the risk becomes if the interfaces is changing, then you don't actually have a need to go ahead and bid on your own name like that or bid on it defensively. I hear you that, you know, marketers are doing this for a reason. But I'm just saying if if the search organization now it is changing because people aren't going to Google search to search for a brand because they're starting something in AI and AI could just take them directly to a website, maybe with a eyes embedded browser or something like that. 26:01 That's a way you're kind of cutting them out of that transaction. 26:05 Speaker 2 Yeah, but, but again, but I'm, I guess my point is like, I don't believe that that's very much of Google's revenue is the point I'm making Like just just just, I'm bidding on my own brand because someone is like they know exactly what they want, but it's just easier to go to Google and type in my brand name than it is to type my brand name into the browser. 26:21 My sense is like, that's probably, and again, I don't have any data to back this up because there is no data on this topic, but my sense is like, that's a very tiny minority of Google's ad revenue. And so there's just not that big of a vulnerability there anyway. But I do believe that there will be ads and agents. 26:38 And I do believe that that opportunity will nonetheless exist if it does exist to any meaningful degree. Because when you're researching some brand and, or you know, you don't want to go to the website, but you want to type in your agent and have it pull up all the different options so you can click out. 26:54 I do believe that that opportunity will exist anyway. And so that advertising opportunity may not be like I'm bidding on my own key. I'm bidding on my own brand, but I'm bidding against somebody else's brand to be represented when that that discovery is happening. And I do believe that that ads will exist in agents. 27:10 So we could, we could talk about that if you want, But like, my sense is the economics always resolve to ads. There's no ability to monetize the agent with a take rate or any other model. It's going to have to be ads if you're monetizing discovery because I don't think going to be able to monetize the transaction for a number of reasons. But if you're monetizing discovery, it always resolves as. 27:25 And so my sense is like, I don't believe branded search is a very big part of Google's business. So I don't think there's much of A vulnerability there. But I also just don't think it's going away. I just think if agents do become the primary source of discovery, I do think that same opportunity will exist in agents. 27:42 Speaker 1 Yeah, Yeah. Just we, we could shift topics here a little bit, but just to kind of put a pin on that conversation. I think that your point taken that, you know, engagement can increase on Google AI surfaces. They still have the opportunity to show ads there. 27:57 They're actually getting more data, more context to show more target ads there. And so as a result, they'll be able to transition from traditional Google search to AI search and show ads. They're all the same. I, I hear you. And that makes sense to me. I guess I'm just a little skeptical mostly because of just how good of a business Google search has been historically and and just how hard it is, you know, if you just look at history to replace such, you know, a stellar business. 28:23 Speaker 2 No, I get it, right. I get it. Like I, I mean, it's like, you know, I'm trying to come up with an analogy, but I can't. But I understand what you're saying. You, you created one of the best businesses of all time. Like, what are the odds that you're going to create the next best business of all time? And like, sure, I get it. And it's hard to make these big transitions. 28:38 And Google is a big company. I mean, even if it's probably more nimble than people give it credit for, it's still a huge company right now. What I'm not saying is that I absolutely believe with total certainty that Google will dominate this next paradigm. I'm not saying that. What I am saying is I don't think it should be taken for granted that it will not execute well right now. 29:03 Could it not execute well? Of course. I just don't think it should be taken for granted that it will not execute well. That's my point. I'm not saying Google is certain to dominate this next era of human compute interactions. I'm not saying it. I'm just saying that I don't think you can say definitively, structurally, Google is very likely to not execute well in this next era. 29:26 That's all I'm saying. Now. What I will say though is maybe just make a point because I am very, very optimistic about the opportunity with chatbot ads. I think this presents like the sweet spot for advertising going back to the point I was making before. You've got monetizable context, then monetize it, monetize against the context, monetize against. 29:46 I'm searching for a new TV. That's a perfect opportunity to serve ads, right? There's a clear commercial intent, but you also have the opportunity to monetize against non monetizable context. I'm searching for random stuff. 30:03 I'm asking for, you know, ideas for, you know, dinner tonight or something. I could show ads against the things that I know that you've shown commercial intent for outside of the chatbot. And I do think the chatbot experience provides for that use the whole Buffalo opportunity that I can monetize when there's context that can be monetized and I can monetize when there's context that can't be monetized. 30:26 I get the opportunity to do both right now. Is that to say that that's going to happen in Gemini? I don't know as you said, Gemini doesn't have ads right now, but I do think that there is a point that goes under appreciated here, which is that if you have a multi turn conversation, you get a lot of opportunity to extract monetizable context, right. 30:44 And so like, if I do a query to whatever, it's ChatGPT, it's AI mode, it's, it's, it's, it's meta AI or whatever the case may be. And you know, it's pretty broad, it's pretty vague. 31:00 It may make sense to not serve an ad there. I don't have enough context. I don't have enough information to really discern like what the best ad Cape could be. But like, that's the start of a conversation. Now I'm going to get replied to and then I'm going to reply to that and I'm going to give more information as I do. And we're going to hone the commercial interests, the commercial intent. 31:16 And I think what you might get to, and I do think this would be a great outcome is better ads and few of them, fewer of them. And so if you got fewer but better ads, they're going to be, you know, they're going to be worth more to the advertiser. They'll pay more for them, but there's actually fewer of them. 31:33 And so the, the, there's, there's sort of like less of that drag on engagement and ads are a drag on engagement. I mean, you see that in, in, in published research, but you see that also just talking to companies like when you increase ad load, engagement goes down. And so wouldn't that be a great outcome, right, If chat bots provide the the opportunity to just get much more better targeted ads and so you need fewer of them? 32:00 Will Ads Play a Role in Chatbots? Yeah, thinking of how that could actually look, if you think about going to kind of ChatGPT, you're in a conversation, you ask ChatGPT, hey, can you help me find steak knives? This this is a real example, actually. 32:17 And I tried this the other week and I wanted chat to help me find steak knives because if you go to Amazon, you search for them. It's just way too many listings, way too much to go through. Amazon's never really been great at that kind of commerce. And so I wanted chat to help me with recommendations. 32:33 However, if it gave me an ad at that point, I wouldn't want, you know, I wouldn't click on that ad. And I'd also then think a little bit suspect about the results that it was giving me because I wanted its input to help me make a purchase decision. And I think This is why, you know, news is talked about trying to monetize the take rate. 32:51 And maybe there is a math problem there with the amount of usage and and cost to service that user. The spending just won't be high enough to to support that. But and I could understand they they wanted to get around that kind of potentially perverse incentive where you have the chat directly recommending something against an advertisement. 33:09 And I might follow up too with chat and say, hey, like, what is this advertisement for? Do you recommend that product? And it's going to have to be honest. At least it should. 33:17 Speaker 2 Yeah. Well, no, but I mean, so like the a modern ad platform, you know, they don't just sort of like randomly throw ads into your feed, right. I mean, these ads are optimized for relevance. And so you know, the at the end of the day, like, So what goes into the auction is the expected value of the the ad, right? 33:33 And that expected value is adjusted by the propensity, the probability of of converting on whatever basis the advertiser is is bidding right? So you're converting for a click or converting for a purchase or converting for whatever. And so you know, these these ads are already relevance optimized, right? 33:49 You know, so like the the idea that like it would just be some random distracting thing. Yes, I agree that would be problematic. That would make you, that would be confusing that, that would, that would, that would make you trust the chatbot less because you're just getting irrelevant content shown to you. 34:05 But the reality is, I mean, it's going to be very relevant. Any modern ad platform will make that as relevant as possible. Now that since there's a bit involved, it might not be the most relevant thing, but like a lot of times, you know, the bit. So the bid could sort of like make up for a lower relevance item, but then it's going to have to compensate for that with a much higher price point, right? 34:26 And so that's, you're going to so like the issue here is I think less about, does this make me question the validity of what you're telling me in the chat? Because maybe the chat would just be nudging me to click the ad. 34:41 But if I, you know, the question is more like, does this ad actually provide even more strength to the, the, the my, my trust in the chat because it's really relevant. And so it's like, OK, wow, this platform is actually really capable of delivering me very relevant, very tailored, very personalized information that meets the needs that I'm, you know, suggesting I have. 35:03 And so sure, like it's going to monetize some of that. Some of the suggestions are going to be monetized. And like, that makes total sense. Now, if you go back to the Amazon setting or the, the, the Amazon issue, you, you, you sort of you, you, you highlighted. 35:22 And by the way, I mean, I agree that that's a problem, but like, you know, anecdotally, I don't know that we can point to like, well, I think this is a problem there for Amazon sucks because Amazon's really good at ads. Like they, you know, it's their $80 billion ad a year ads business. But like, The thing is, here's what I think that's ultimately going to look like. It's Amazon then gets the opportunity to buy my attention in that moment by inserting an ad there. 35:44 And in that way Amazon wins, Open AI wins and the consumer wins because Amazon again is really good at tailoring ads. And so if Amazon is able to say, look, I am going to shut off my platform to these third party agents unless I'm able to insert ads into the chatbot output. 36:02 And if I can, then you get access to my whole catalog. And if I can't, I'm shutting you out, right? I think then everyone wins. That's much more incentive aligned. And if you say, look, no, I need to take a take rate, The math problem isn't like it costs money to operate this and I have to make money. 36:17 The math problems really an optimization problem because what take rate do you pick? 5%, four percent? Why? Any take rate that you pick arbitrarily is going to, it's going to shut some some retailers out because they can't afford to take that margin hit and it's going to under price what other retailers would have been willing to pay. 36:38 Just by definition, any take rate that you pick arbitrarily will disqualify some retailers from participating altogether because they can't support that take rate and it will under monetize retailers that would have been paid more. Wouldn't you rather have every retailer be able to tell you exactly how much they are willing to spend for that, for that, that moment of the user's attention, for that exposure to the user and then just decide which of these is the most relevant and how do I adjust it by that amount of money that the that this retailer is willing to spend. 37:09 And then you've got an auction and then you've got ads. I think that is ultimately how this is going to resolve. The take rate thing will not scale. There is first of all, there's a mismatch between the retailer's incentives and the plot and the agents incentives because some retailers just don't want to be intermediated in that way. 37:25 And then ultimately, you're going to be under monetizing. If you want to maximize the revenue opportunity, you have to open this up to an auction so that every retailer can bid their true value. 37:35 Speaker 1 Yeah, I think on the take rate thing, I mean Meta has a unique we'll call it trust problem with definitely not being able to just you know get Muse usage initially while talking about ads and we'll see what happens 100% yeah down the line. 37:49 Agentic Commerce But on, on the take rate point, why would an other retailers just raise their prices if it's being purchased through agentic commerce or something like that? You know, I think there's some studies that that show that, you know, once you know, credit cards are introduced to an area, the interchange fees essentially just result in, you know, prices going up. 38:09 Speaker 2 So I'm gonna pay more to use the agent to buy than if I went directly to the retail platform. That's the value proposition I don't know. 38:18 Speaker 1 You're paying for convenience. 38:20 Speaker 2 I pay for convenient. How convenient is shopping on Amazon? I go and I click one, click buy. I go buy it, and then I get the stuff for three hours. Amazon is the most convenient shopping experience has ever been introduced to human beings. There is nothing more convenient than Amazon. I think what people are think about when they say convenience, I mean everything is available in the agent. 38:36 I can access any retailer, but you can't. You can't access Amazon right now. So that already isn't the case. Now the question is, do I go to Amazon directly or do I stay in the agent? If you're telling me it's going to cost more to buy through the agent, of course, I'm already not buying it from the agent if I need to go to Amazon because it's not available there. I mean, I think the reality is like you're going to get into this race to the bottom. 38:53 There's going to be many agents. We're thinking about Muse right now because Muse is the dominant agent, but there's going to be a lot of agents at some point in a year's time, there's going to be many agents and they're going to compete for users. And if they have to compete for users, you can't say, oh, and by the way, every single you per every single purchase you make, the take rate that we demand from the retailer will be passed on to you because what's going to happen? 39:11 You're going to get a race to the bottom. What's happens? You get to race to the bottom. You don't monetize the take rate, you monetize with an auction. 39:18 Speaker 1 Yeah, I, I don't actually disagree with any of that. In terms of kind of the agenda commerce attempt so far at least the current experience, it's not really great. I do wonder if it, because the problem with a lot of these arguments is they're very dependent on hypothetically how things can look in the future. 39:37 And everyone will say, well, right now it's the worst it's ever going to be ever. And So what could this look like in the future where you do have an AI agent that is capable of actually scraping the web instead of, you know, just pulling the last few searches. And if you could be in an environment where maybe, you know, Amazon continues to gate, but then, you know, Walmart thought it'd lead them to a competitive advantage if they let, you know, AI agent surface their results there. 40:03 And so I don't think, you know, everyone is going to be able to be in this walled garden. I think some people are going to try to take advantage of it. I think we saw that. Ready a little bit with, you know, Expedia kind of partnering with Muse at launch and I think some companies are going to try to use this as a way to kind of re reposition their businesses, whereas others like Amazon, who are in a great position right now and could kind of only be LED worse off as far as I could see it from AI agents, they're going to try to hunker down. 40:32 Speaker 2 Well, worse off by third party AI agents. Keep in mind Rufus generated $12 billion in incremental annualized GMV for Amazon last year. So they're worse off if third party, if they just said, look, we're going to take the wall. 40:45 Speaker 1 Is that so? Can you explain that? Because I I've seen Rufus before in the app, but it's, it's been lackluster, we'll say. 40:52 Speaker 2 I I get it man, but like I said, anecdotes are will only take you so far like you have to understand. 40:58 Speaker 1 Like which is how does it work? Is it people going to Rufus to ask for recommendations or? 41:03 Speaker 2 Recommendations. But they also have ads in Rufus now. But The thing is like, I just think I here's, here's why I, I and, and, and I know I responded to your tweet and, and maybe it was a little brusque, let's say, but like the point I was making is like, anecdotes will only take you so far in this conversation. 41:20 I really do think you need to bring it back to history of consumer behavior over time and 1st principles on the economic incentives because like I hear this argument a lot, like Rufus sucks. I use it, it sucks. Well, it did $12 billion in, I mean, it generated $12 billion in incremental annualized GMV for Amazon. 41:36 So like you maybe are not the median buyer. And it's like, well, I used Muse and it was great because it got me, you know, it rerouted my flight and you know, and that saved me a lot of time. It's like, sure, go talk to the median consumer. 41:53 How often are they flying? It's like, I think when you think about agent to commerce, I do think it makes sense to anchor this to the median consumer and to think about the economic incentives of the platforms and the retailers. And so I think that's like that's going to that's going to take you further and be more instructive than like, oh, I bought this thing one time and it was really convenient. 42:13 But to the Walmart point, it's a good point. Look, Walmart's participating, they're partnering with Muse. Amazon is not Walmart did participate in Instant Checkout, partnered with Open AI for instant checkout. Amazon did not Walmart. Walmart also disengaged from Instant checkout and then the VP of product had an interview with Wired where he basically badmouthed the instant checkout product. 42:35 These are very different companies. Walmart did $6.4 billion in advertising revenue last fiscal year. Amazon did 69 billion AD revenue represents 88% of Amazon's net income last year, represent 29% from Walmart, right? 42:52 So these things matter, right? Like advertising is very important to Walmart. And like the executives have said, this is like this, this represents like the preponderance of our profit growth. It is very important and they can't be intermediated for that reason because they need this profit center, but they also have a much lower share of the revenue being driven by E com. 43:10 And I think they benefit disproportionately by just generating more ecom transactions. Is if they if if agenta commerce is able to do that third party, then maybe that's more it's more beneficial to them than it would be to Walmart. But at the same time, they have their own agent now. It's not it's not like a fully fledged trans checkout experience like it is with Amazon. 43:29 It's just more of a discovery tool. But they've also said like, yeah, there are our on platform agent delivers a lot of value to us. The AOV is higher when people use it and the card sizes are larger. And so they also have their own agent that I think they would prefer that people use. My sense is like Walmart's a very different company than Amazon. 43:45 And I don't know that you could look to Walmart as being the bellwether for whether this takes off. But I just do think like, if we, of course, we're considering hypotheticals and we can't see the future, right? But like, I do think we could probably all agree that Muse won't be the only agent that exists, the only like agent from a big scaled consumer tech platform that exists in a year's time. 44:06 It won't be certainly won't be the only agent that exists that wants user engagement. I mean if you look at the full court press that Met is doing right now, it is unprecedented. I've never seen them do anything of that scale threads wasn't even this big. I mean, so they are very, they are they very much they're very sort of like earnestly committed to getting a lot of adoption now. 44:25 And I think that portends a sort of like knife fight for agent share of time spent. And so I do think you've got to kind of think about like, well, when the space gets more competitive, how can they possibly monetize because the transaction fee just won't be sustainable. 44:44 It won't competitively sustainable. 44:47 Subscriptions vs Ads for AI Agents What about a subscription? 44:49 Speaker 2 Subscriptions are great, but like you know, then you just eliminate 90% of the Tam. I mean, you know, Meta, look at the Meta's got a subscription in Europe, in the EUI mean Meta's got multiple subscriptions now, but like their core social media product had a subscription in the EU got like sub 1% uptake. Now, what people said about that was that like, well, OK, fine, but it was a free product first. 45:08 They instituted the subscription as a result of like the EMA. And so there's relatively reason that they introduced this, but you're just going to eliminate a lot of the Tam and and it's going to be the it's going to be people that are just they have the disposable income to pay and the willingness to pay. And there's just a lot of people that don't have that. I mean, demonstrably, I mean, there's no subscription for social media for a reason. 45:26 I mean, Cheryl said this years ago. Like, yeah, we could pursue the subscription model, but ADS is superior. 45:32 Speaker 1 Yeah, I, I'm, I'm kind of in a in agreement on that. I mean, you have advertising, sure that could increase the Tam on the other side of that. I don't know. I just again, you know, relying on my own kind of personal experience and talking to people. I think that some people, if you are talking about having like an AI agent run your life, there is kind of a little bit of more trust issues. 45:56 And maybe that's just me pushing that on. But if you are thinking about, you know, how they're ultimately monetizing the product, You know, if you're in ChatGPT and it's throwing you ads, that kind of seems innocuous. But once you get to the point where you have like an AI agent really controlling more and more for you, I think subscriptions might become more mainstream. 46:13 I think we might see Apple Siri could potentially be the first one to really roll out an AI subscription. And they'd be in a pretty good place to do that because they could kind of force the whole app ecosystem to play along. So I think that could kind of be an interesting way to go. I don't know if you have any reactions to that. 46:30 Speaker 2 Well, I mean, you know, realistically, how many subscriptions would people pay for that? And you're really anchoring yourself to a single agent, right? And so the problem with the subscription again, is like if the monetization is better with ads, and it tends to be, I mean, you just get higher Arpus with ads when done well, you're always going to have someone undercut you on the subscription price and then subsidized with ads. 46:54 And then you get in this race to the bottom and then you end up giving it away for free because you want to maximize Tam and this and then monetize it with ads where the ARPU is already better. So it's just like, I think the economic, you're always going to butt up against the economics of this. Now, now the other thing is like, if you're talking about the segment of user that lets agents run their lives, sure, they're going to get a lot of value out of the agent and they're going to be willing to pay a subscription. 47:20 Don't show them ads. Chat CBD doesn't show ads to subscribers that's fine what percentage of users is that now we're talking about the intersection of people that can pay and people that are willing to surrender emails to good friends to the agent. 47:36 I think that's a very tiny percentage of the consumer base. So I think I just try to think about this from a Tam perspective. And then when we start excluding segments like who are we excluding? And I do think that the agentic commerce discourse is dominated by people on the coasts who, A, are way more willing to allow agents to penetrate their lives than the median consumer and B, are way more willing to pay for that privilege than the median consumer. 48:13 And I do, I do think that it might skew how people how, how, how, how popular, how popular people believe agent and commerce can be just as just as a as a function of as a function of the, the, the, the, the volume of that perspective. 48:33 Speaker 1 Yeah, And I, I think if what we've seen with Netflix is just how limiting subscriptions can be, as they themselves have been, eventually, you know, capitulated to advertising. 48:42 Speaker 2 I mean, Netflix is going fast. They have to go fast. They will go fast. It's a matter of time. I mean, they, they, they are, I mean, especially now, I mean they're, they're, I think in an extremely vulnerable state. I mean, I think the problem is they were so successful, they forced everyone else to go fast and they got on our cut. I genuinely think that's the problem. 48:58 They were, they were a victim of their own success. They were so successful that they, no one else could compete on the basis of content. Therefore they couldn't charge enough to sustain the service. Therefore they had to go fast. And then Netflix got undercut on price. And I think as a result of their success, they're going to have to go fast. 49:17 Speaker 1 When you say they got undercut on price, you're talking about other ad tier supported subscription services. 49:24 Speaker 2 Or just very popular fast services to be Roku? 49:28 Speaker 1 Yeah, got it. So kind of kind of just kind of putting a pin on on the rest of this AI agenda kind of discussion here. So overall, your take is that there's probably going to be multiple AI agents. Everyone is going to best they can have their own walled gardens, which is going to make the utility of AI agents a lot less than what most people think. 49:51 You drawn to question the monetization model of these AI agents because you don't think a take rate is ultimately going to make sense and you think that they're going to all ultimately have to rely on advertising. But even then, the ultimate use is kind of gonna be a little bit dubious because they're not necessarily gonna be able to access a lot of these different retailers. 50:09 Everyone's gonna be playing different games. And so this is gonna really kind of hinge the ultimate adoption and utility of AI agents. 50:18 AI Agents vs OTAs Yeah. But for certain categories, I think that's all true for E com, but I think some categories, there are properties of categories that make it especially well suited for agent of commerce. Travel is 1. I think travel will be done agentically for the most part. It already kind of is. That's what an OTA is, online travel agency. It's an agent, right? 50:34 Travel has all the pipes. They've got all of the infrastructure set up to allow agents to access inventory, purchase inventory on a user's behalf. All of that exists, right? And hotels and airlines will never get intermediated from the customer. Why? Because you got to show up to the hotel. 50:50 You got to show up to the plane and you're going to be very heavily incentivized to let them know who you are with your loyalty number because you want to get the rewards for that, right? So that loyalty system creates this sort of like unbreakable bond between the service provider, the retailer, the hotel or the airline and the customer, but. 51:12 Speaker 1 In bookings case the the loyalties more commonly at booking than what? 51:15 Speaker 2 Well, yeah, it could, yeah. With booking, but it's both, right. I mean, because I'm getting so, so I have the the booking points or whatever, but also I get the, the, the airline loyalty points. Now the issue with the OTAs though, is they take a rake. They take a rake on, on it's, it's less so on on flights. It's it's, it's diminished. 51:32 But, but with, with hotels, it would be up to 15%. Now that's all essentially passed along to the consumer. Now, there's there's nuance here because there's different models, but like, if that goes like that's it's the economics are more aligned to use an agent to do that. If you're not having to pay the middleman to just do discovery, if the agent is doing what the middleman was doing, which is discovery and arguably doing it better, then you know, there's there's a total economic alignment and you don't have to pay for it. 51:57 There's total economic, you pay much less than 50%. There's total economic alignment to use the agent between the ultimate service provider retailer and the consumer. Now, the other thing here is that fulfillment's not a wedge. Fulfillment is not something that costs a lot of money because what's fulfillment? What's fulfillment for an airline? 52:12 It's getting the plane off the ground. You were going to do that anyway, whether that consumer showed up or not. It's the fulfillment is the consumer has to get to the airport and then the plane flies from A to B. It's not Amazon's got to get you the stuff in three hours. And so they have to invest a tremendous amount of money in that infrastructure, right? 52:27 So if you have to replace true fulfillment infrastructure, then that is an impossible hurdle to clear as an agent. But if all you have to do is just get the consumer to the place they have to be and the airline takes over the rest or the hotel takes over the rest because the hotel exists whether the consumer shows up or not, then there's no fulfillment hurdle to clear, right? 52:45 So I think there are properties of certain categories that make it well suited for agent e-commerce. I just don't think e-commerce is 1. 52:51 Speaker 1 Yeah, I, I mean by and large I agree with you on the, the OTA take. I, I don't think it would be quite as easy to, to displace, you know, someone like booking on the hotel side. Airlines is a totally different business for them. But if you are looking at kind of just general pricing between OTA's versus going direct, there's still a good amount of times where bookings actually cheaper because they contract out for bulk rooms ahead of time. 53:15 And so that allows them to actually have a lower price. But, you know, even if it is, you know, one out of five or one out of 10 rooms or something like that, that it's cheaper to go direct, that definitely could still be a headwind for them. And, you know, they themselves, they do still have their own sponsored advertising business. So if people are, you know, visiting their site, that's, you know, another hit for them. 53:34 Speaker 2 Well, but, but, but keep in mind like so, yeah. So I mean, that's what I was talking about with the nuance. Like so you either act as the agent or you you, you get the bulk and and you you sell it. 53:43 Speaker 1 Yeah. Merchant first agent, Yeah. 53:45 Speaker 2 But The thing is like, they don't pay upfront for the rooms, they don't pay for the guarantee, right? And so the fact is if like, if they're and if they're end up like if they're under pricing, that if they're under pricing it, then the retailer is worse off, right? So what you'd rather do is like create this kind of spot market where people are just buying as it's priced, right? 54:02 And so I think that if the discovery solves for that, if the agent solves for that, because what Booking is doing is it's basically guaranteeing a lot of visibility, right? Because a lot of people use Booking and for a hotel that's very valuable because there's a lot of hotels people don't know about your hotel, you're some boutique hotel. It's different for airlines because there's a handful of major airlines. 54:18 Everyone knows the airlines and they have, they've already got like loyalty points. There's many different hotel options. You don't, you know, a priori, you don't know about the hotel if you've never been to the city, right? And so it's going to take a lot of, you know, a research on your part. If you go to booking, they consolidate everything and make it much easier to discover. But the agent can do that. 54:34 So I think like, and booking has got a, they've got like a nascent, you know, it's not sort of structurally important, but they do have a display business. So I'm sorry, a search business. So, you know, I do think, and the other thing is like booking spends a lot of money on ads. Well, someone pays for that. 54:50 Like that is money that comes from somewhere. So I mean, I think if there's no need for the discovery surface to buy ads to essentially, you know, then reconnect users at the at the sort of like higher price point than what they acquired them for because it all gets consolidated in the agent. 55:07 I think that ends up just all those savings probably go to the consumer or the surplus goes to the hotel and everyone's kind of better and those two parties are better off. 55:15 Speaker 1 Yeah, it kind of makes sense in theory. I just, I don't know why that wouldn't have already happened even with just like a Google search world. Because I get, you know, now you have agents that could be a little bit more proactive searching and all of that. But this idea that all of a sudden that's going to add so much more liquidity to the market that you don't need someone to kind of, you know, purchase and bulk your rooms, I don't know why it just wouldn't have already happened. 55:38 Speaker 2 Well, I mean, because I, I, so I mean, first of all, these, these, look, I mean, we, you know, we talk about these things in the abstract. And I certainly tend to talk about these things at the limit because I just think that's, it's more interesting to think about like what the end state's going to be. But brands are sticky, consumer habits are sticky. 55:55 These things take a long time to erode, right? And so I mean, with that caveat, you know, acknowledged, why wouldn't it have happened already? Because natural language search just allows you to add in a lot more detail, right? Like a keyword, you know, Google search is still short form. 56:11 It's very short form. So like hotel in New Orleans is a lot different than I want a hotel in the French Quarter. It should be, I was in New Orleans recently. It should be, you know, whatever less than a 20 minute walk to that place that makes the beignets. 56:26 And I might actually just say that place that makes the beignets, it's got it's got to have a gym and I want to make sure that I have a room with a courtyard view. That's that could be a prompt that I type in just stream of consciousness and that could find, you know, appropriate hotel rooms in search. 56:42 You can't you still can't really do that. I do think this longer form format is kind of a why now, you know, a wine, a piece of wine now support. But again, to your point, consumer habits are sticky. Brands are durable. 56:59 They don't erode overnight. I'm not saying this all at you know, I, I talk about things at the limit, but it's not like you snap your fingers and booking.com goes away. Sure. 57:08 Speaker 1 No, I think that's a good insight. Why don't you want to apply that to e-commerce where now maybe the long tail of e-commerce is a lot easier to find and you could find a product that's more suited for you through an AI agent. And you know, that's kind of more of the the traditional retailers that have a more limited skew are kind of at risk. 57:27 Speaker 2 No, I, I do want to apply that to e-commerce and I think that will be a use case for it. It already, I'm sure it already is. But my my point is the retailer has an economic incentive misalignment with the agent then doing that transaction on the user's behalf because they need to maintain that user connection. 57:43 Now the airline, if I find the perfect hotel room through an agent and I transact through the agent, I still want to give the hotel my loyalty number. I'm still going to show up to the hotel in person. I'm still going to have a direct relationship with the hotel. If I have my agent purchase on my behalf and the retailer has an advertising business that's very important to it. 58:04 The hotel doesn't have an advertising business that is getting sort of competed away, right? So there's, these are different, there's different properties here. And the, again, the fulfillment piece, the, the retailer wants to own the first party relationship with me because they want to upsell me in numerous ways on platform later with remarketing with discounts later on. 58:26 They want to maintain the relationship with me. Plus, they've invested all this money in using fulfillment as a competitive advantage. The hotel's not doing that. I show up to the hotel, I get to the hotel, I fulfill myself to the hotel. So there are differences here. First of all, the hotel never runs the risk of losing the first party relationship with me because I show up and I want to give them my loyalty number because I'm going to get rewarded for that. 58:46 But if the agent's buying on my behalf from the retailer, the retailer doesn't get that information. They don't have that. They don't necessarily get to aggregate that information to me, right, to my, to my profile. And they don't get the opportunity to upsell me to, to show, you know, to send me discounts later on to, to, to maintain that sense of identity, to spot me elsewhere on the Internet. 59:05 So there's just there's, there's, there's fundamental differences between these categories and these these service providers and retailers that I think make one suitable for agent of commerce and one unsuitable for agent of commerce. 59:16 Speaker 1 Yeah, I think that all makes sense. 59:18 How Apple Could Monetize AI Just maybe switching gears a little bit to kind of we briefly touched on, I briefly alluded to Apple's Siri and I was wondering how that could potentially change the app ecosystem and maybe also the way people kind of advertise out in the future on mobile apps if Apple would ever potentially become more involved. 59:39 There is kind of a second corollary question. 59:42 Speaker 2 Well, I mean, it's really interesting, right? So I mean, when Apple first announced the Apple Intelligence, you know, they talked about all the ways that Siri would be this kind of like hub for discovering content and doing things, taking actions, right? 59:58 And they had this app intense framework. The app intense framework is old. It predates Apple Intelligence, but they had integrated that into Siri in a lot of like interesting ways or had at least alluded to the integration that would come back in when Apple Intelligence was first announced. 1:00:15 I don't want to say it, it was WWC 2024, WWC 2024, I think. And then? Nothing ever happened, right? So then at WWC this year they announced even more functionality for app intense. And I think it actually will become a much bigger part of the discovery process because or at least the the the interaction process because you you can then use the chat interface or the voice interface, you know, seriously the chatbot now to do things in these apps that otherwise you would have had to do directly, right. 1:00:45 So I do think it's going to absorb some of that engagement time, interaction time. Now, there's certain categories where this just will never apply, right? Games, you're not going to play a game through Siri. You know, you're not going to play Candy Crush through Siri. So you're always going to go to Candy Crush. 1:01:00 I think entertainment apps, you know, Netflix, same like you're going to go to Netflix, but a lot of like utility apps, you know, there may be no reason to actually open the app. And so like, if you think about an Uber, right, like, so Uber is actually, you know, integrating with Muse, right? 1:01:15 You can order an Uber through Muse. They're also like integrating with with Siri. And like that, to me seems like one of those cases where if they don't do it, a competitor will. And it'll just be more convenient to use Siri to do those things, to take those actions than it is to go into the app. 1:01:35 And that's a real vulnerability to them right now. Uber does have an important ads business to protect, but maybe that the calculus just doesn't work in that favor for Uber. It's like, well, OK, if we're not where, if we're not where consumer time spent aggregates, we're going to be competed away, we're going to be forgotten and we just can't afford that, right. 1:01:54 So maybe like a lot of these services probably don't have the leverage that like an Amazon would have with agentic commerce. So I do think Siri will end up taking up a lot more of the engagement time spent, but it's going to be disproportionate based on the different categories and like what, what functionality those categories fulfill. 1:02:16 Speaker 1 And do you think Apple is going to ultimately try to monetize this with subscriptions or do they dare get into advertising? 1:02:24 Speaker 2 Advertising in Siri. 1:02:26 Speaker 1 Yeah. 1:02:29 Speaker 2 No, I don't think I might. Just my intuition tells me they wouldn't do that. I do think, I think, I do think Apple has, has, I, I do think Apple's advertising ambitions are larger than they have been in recent history. 1:02:49 And I do think potentially with the leadership change they, they see an opportunity to pursue those more aggressively. I mean, you see this with the adoption of ads in, in Apple Maps. Now that predates Turnus, but still, I mean, I think they've, they've, they've also made some hints that like they want to make some, they want to make more money from the App Store. 1:03:09 And I just don't know how they would do that outside of advertising because they can't increase the Commission. It's just, you know, the commission's not a threat. So my sense is like, advertising is probably the only path to doing that. I could potentially see a subscription tier for Siri. I mean, I think that would be consistent with all these other chat bots. 1:03:25 So like, why not do that? I also think what Apple is in a good position to do is to create. So my sense is like if Siri became just kind of like generic chatbot interface, like connect Siri's on device, it's available to you already and we can actually just have this connect to any number of provider models or provider APIs. 1:03:55 Then you have a pretty good opportunity to do like, well, default model provider status like you do with search. And then like take a cut of if people say, well, I need to upgrade because I'm hitting my API limits. I think that with search on Safari, I mean, I think that could be an interesting model for them, you know, but I think, you know, the other area where I think they could introduce monetization here is private cloud compute, right? 1:04:18 So that's their sort of like privacy centric way for compute for doing processing off device like so they've got, you know, they've got the whole, you know, they brought the on device model with with the latest a pretty capable on device model with the latest iOS update, but it's still like constrained relative to like the largest frontier models. 1:04:38 And so, you know, if there's some task that requires A heavier model, you've got to go off device. And if you want to go off device, but you know, use, you know, iOS, you've got to go private cloud compute. I think they could monetize that. I think that could be this compute interface where they say, look, if, because, because what they do now is they've got this pop up that says if this app is using a third party API service or, sorry, a third party AIAPI service where the data flows off device, you've got it, you've got to get consent. 1:05:05 It's it's, I've called it AI tracking transparency and it, it operates very similarly. And but if you use private cloud compute, you don't have to do that because it's the privacy, it's Apple's privacy environment. I think they can monetize that. So I think they have a number of pathways to monetizing like AI in the iOS ecosystem. I don't, I, I think subscriptions make sense. 1:05:22 I think they they should do that, but I also think they've got other avenues for that. 1:05:27 Speaker 1 Yeah, yeah. I mean, it's like, on the one hand, I feel like advertising for them in from like a business perspective, it'd kind of be a no brainer. Except at the same time they just can't do it because of their culture and their whole campaign against Facebook and Apple privacy and and all of that. So it's really kind of impaired their ability to to roll out advertising. 1:05:45 Speaker 2 Yeah. Well, yeah, I mean, I think like we're seeing in Europe that, you know, their their ability to sort of maintain ATT has been diminished, right. And so they've had to make a lot of concessions around what they how they talk about their own opt in prompts for their own ads targeting and unifying that with ATT prompt. 1:06:08 My sense is, you know, privacy was probably like a good PR campaign for them for a while. I don't know if you just see people fully opting in to, you know, using chat bots to like manage or to, to process like very sensitive information, which I, I do think probably. 1:06:28 So like there's, there's two, there's kind of like 2, there's two different extremes here. There's two sort of like extremes that are on different scales or like different axes. 1 extreme is like I use an agent to run my life. I think that's like a coastal elite usage extreme. 1:06:45 I don't think the medium consumer goes there. There's another extreme is like, yeah, I use ChatGPT and I'm willing to upload my blood test to it to ask it if it sees anything weird. I think a lot of consumers go there and it's it's just irrespective if you're on ChatGPT Pro or ChatGPT free, I think people just end up uploading a lot of really sensitive information to like the very easy to use chat bot because there's no real sensitivity to that. 1:07:11 And so my sense is like, if, if that does become the norm, then like the privacy PR campaign kind of loses a lot of it's a lot of it's economic power. 1:07:24 Speaker 1 Yeah, yeah. But then the other side of it is I don't know how much people really care about data privacy and how much it was just kind of an advertising point versus a real preference, kind of, you know, stated versus real consumer preference, especially as we see how much data people pour into, you know, Meadows, Muse out of all apps. 1:07:42 Speaker 2 No, no, I agree. I agree. I agree Meta was #1 for more than more than two weeks. I mean, like, you know, if people had any reservations about, you know, interfacing with, with Meta products or giving data to Meta products, they, they certainly didn't exhibit that right. So I, I do, I do agree 100%. 1:07:57 I think it was, it was a good PR campaign. I don't know. Yeah, you maybe you're right that it it never really was that valuable from a commercial perspective. But I think it becomes even less so when when people are just, you know, dumping their medical records and like legal contracts in a ChatGPT free and asking for summaries. 1:08:15 Speaker 1 Yeah, yeah, yeah. No, I do wonder about them because I just did this report on Apple Lovin and you know, disclaimer for the audience you are an advisor to to Unity. So you're a little restricted on, on what you could say. But Apple is very clearly the one competitor there that it just seems like they could crush everyone in the app, mobile app advertising environment, maybe barring some antitrust legislation. 1:08:38 But they they clearly could have a much bigger their business there if they really wanted to. And it seems like it's something they've kind of willingly seated and any kind of, you know, mobile app business, you kind of have to hope that that continues to be the case, at least in my opinion. 1:08:55 Speaker 2 Well, they're, yeah, they're closest to the metal. I mean, I, I always kind of think about it from the perspective of like, you know, they know what apps are. And so I mean like if, if, if, if they are able to maintain this positioning of like anything we do is by default privacy safe simply because we're the first party. 1:09:16 Like we're, we have a first party relationship with the consumer because they're on our device and they're in our, in their, in our OS. Then they know what apps you have installed. They know what apps that you spend money in. I mean, they, you know, they know what subscriptions you have active. And now like they do a lot of stuff that is very privacy safe with the way that they aggregate this data and put users into targeting groups. 1:09:35 They don't really target against individual profiles, but nonetheless, they have all that. And so if they, you know, said, look, you want to why don't, why don't, why don't we serve your ads and your game? Like why don't we do that? 1:09:51 And by the way, by default you have our mediation SDK integrated into your app because that that's just because it because it's not even like an app SDK. It's just some functionality that it's an API that we make available from from iOS. They could spin up a third party ad network in in pretty quickly. 1:10:09 So, yeah, I mean, they just have such a proximity to the to the metal, to the OS to also the metal and the OS that, that they would have just a structural advantage and and they have access to this wealth of data about everything that you do on your device. 1:10:24 When Will Ad Model Gains Plateau? Yeah, yeah. And it kind of just as we wrap up here, I kind of did have one question I've always been wondering, which is that, you know, we look at Meta, they are clearly incredible at return on ad spend for advertisers and you know Google is really great too. 1:10:41 My, my question is at what point would we know that, you know, the gains in the ad model are starting to slow down because you know, with AI everything is accelerated and you know, they're, they're both of the company's revenues growth have accelerated recently too. 1:10:56 And it doesn't seem like, you know, their ad model targeting or measurements are, are slowing down in, in terms of gains anytime soon. But at some point you got to figure that's going to plateau. And so is there any way to know ahead of time when that will start to potentially plateau or is it something you just have to continue to incrementally monitor? 1:11:16 Speaker 2 That's a really good question. I so there's a couple things like 1 is, is, is, is. So first of all, I think it's really important to recognize, and I think this is an underappreciated point that that Meta and Google, they're not just like, you know, it's like it's, it's one, it's they are like very effective expert operators of ads platforms, but they also are at like the forefront of AI research in a lot of ways. 1:11:44 Certainly in like the Rexis area, you know, Google introduced the world to transformer, the transformer architecture, right? And so you know, these, these these companies, they, they, they publish like like bleeding edge research that ends up like moving, you know, these sort of like the, the entire field of AI forward when it happens. 1:12:07 And so they, they are going to be ahead of like most any other company in terms of like applying that to their specific revenue engines, which is, you know, our ads based. And so like Meta is constantly talking about the improvements that are, that are derived from GEM, you know, from Andromeda, from Lattice, from all their like ad ranking infrastructure. 1:12:34 And you know, my sense is like they're, they're moving the frontier here in these like very fundamental ways. They just published A blog post yesterday about their use of semantic IDs to do kind of like next ad prediction in AD for AD ranking purposes. And like these are very, very frontier areas of research. 1:12:50 So my sense is like, and, and, and the reason that they say these things are important is because they're tying them to these model architectures and particularly like LLM architectures that benefit from scaling laws. And so they can say that look, look, the more compute and the more, the more data we throw into this, the more performance we get because we have like predictable, predictable improvements. 1:13:12 And so from that perspective, I don't know that there is a foreseeable end to that. But I do think like when would you think that the music's about to stop? I think it's, I would tie this to like examples of ad platforms where it has and they tend to blame macro factors, you know, demographic factors. 1:13:33 They start talking about all these things that they don't really have control over and in in a way making excuses. I think when you start hearing that from the biggest ad platforms, then you might expect, OK, maybe the next quarter's not going to be very good. 1:13:46 Speaker 1 And I this is at least where my mind is going right now. And you know, you could say if, if you can't talk about this because of your unity relationship. But app lovin in the last quarter they did misguidance and they kind of had slowing revenue growth for the first time and or at least misguidance for the first time. 1:14:04 And they kind of attributed that to not having enough model gains for the quarter. That was one of the the big attributions they had to it, which kind of was leading me to this question where I'm wondering, you know, at what point do you really, really know when the model, you know, stops producing gains And they, you know, said later on, oh, we've made more gains and made it up this quarter. 1:14:23 So it's not really just an app lovin specific question, but that just goes to show how quickly you know these things can potentially manifest in the real world. 1:14:31 Speaker 2 Yeah. And I mean, look, I'm not like under a gag order or anything like that, but I mean, yeah, it was, I mean, so I think like it's, there's also like a, a function of like how much, you know, money you're actually investing in this research, right? 1:14:47 And so the, the, you know, looking at Meta and looking at Google and thinking about like the CapEx expenditure, but also just the size of these teams. And, you know, you, you, you see the, the, the quality of the talent that they attract. I mean, this is very expensive proposition. And this is, and I'm not speaking to Apple Evan specifically, but like, because there's two, there's two separate things, right? 1:15:07 To me, Apple Evan misguidance and you know, the revenue engine is still gaming, but also I think there was a sense that they oversold the near term impact of E com and so like that. I think both of those things had an impact on have had an impact on the stock. 1:15:26 But but you know, like when you think about like the enormous amounts of money that you have to pour into these research efforts, like it is a very, very rarefied group of companies that can actually do that at scale in a way that is dependable. And you talk about bending, you know, these bending the scaling laws and benefiting from benefiting from the scaling laws and the sort of reliability and the predictability of more data, more computing because it's better performance. 1:15:49 I just think that there's very few companies that actually are at sufficient size and scale to reap those benefits. 1:15:59 Speaker 1 And and just because you mentioned ecom, is that something that could potentially create even more data for these models? So then you not only have more opportunity to select from more ads to better match them to a user, but then that in of itself it's generating more content and more data. 1:16:16 And so it's just continuing to give it like more runway to continue to reiterate the AD model that acts on 2.0 model they have. 1:16:23 Speaker 2 Yeah, sure. I mean, more data is better, more data is better. And you know, like, I think there, there's been research that, you know, these, these these categories are not these strict silos that you can actually borrowed. You can borrow, you can borrow essentially learnings representations, certainly, but learnings from one category to another and and benefit from that. 1:16:42 So like, it's not just that, you know, more E com data equals better performance than E com. It could be more E com data equals better performance across the board. So sure. I mean, I think just more data is generally better when you talk about these kind of. 1:16:52 Speaker 1 Models, right, right. And have you heard anything on how their E com push has gone so far with advertisers anything? 1:17:00 Speaker 2 No, I mean, you know, like when you, I think so when you talk to the advertisers that use the product, they love it. They think it's great. Like they, they, they can scale with it and they're seeing great results. I think, you know, one of the issues potentially with the product, and I talked about this more than a year ago. 1:17:16 I mean, this isn't like some new revelation I wrote, I wrote this piece probably a year and a half ago, maybe maybe going on two years ago. I think it might have been November 2024. It was it was app lovin e-commerce bull and bear theses or something like that. And, and one of my bear theses was like, look, they have structured this pilot because it was back when the program was in pilot to, to sort of like target the, the, the, the advertise, the E com advertisers that are best positioned to see success in this particular placement. 1:17:50 And that is E com advertisers that were spending a substantial amount of money as a rule to be admitted into the pilot on Facebook ads. Now, what do you get if you're, if you're spending like some substantial amount of money on Facebook ads, what is probably also true about you? 1:18:05 What's also probably true about you is that you've got you've got robust measurement infrastructure that can accommodate a new channel, you've got creative production infrastructure that can adapt to new placements and and optimize for new placements. 1:18:22 And so, and you've got the ability to do incrementality testing. And so like, if those are true for an an E com is is interesting because as a category, it's driven by the very long tail of advertisers. Gaming is more driven by the the the the largest spenders, largest studios, right. 1:18:37 And so in order to be successful with an E com product, you know, and look and look at Facebook for this, you need to be able to support that SMB because there are a lot of them, right? And so my sense is like, if you can't provide creative for the advertiser and they have no capacity to make creative for a totally wholly conceptually new placement, you're not going to be able to on board them. 1:19:03 And so that that may have been an issue there. But again, I articulated that in November 2024, I said, look, this is, you know, we'd have to be sure that this isn't the case if you wanted to be optimistic about this channel. 1:19:14 Speaker 1 Right. Yeah, the different ad unit. Similar thing existed with reels, but in that case Meta was able to get people to create the new content. Whether or not they could do the same for a relatively smaller channel is an open question. Maybe offsetting that though as you have more generative AI creation tools to help you create ads. 1:19:30 Speaker 2 Well, but that's the thing, because you need video for that. And I don't know that those tools are there yet. I mean, they're not there yet unless you're willing to, to, to invest substantial amounts of money in building those tools. Like you look at like the leader here is Byte Dance. Like they, they, I mean, if you look at the tools they have for, for Gen. 1:19:45 AI video creation, they're, they're incredible. I mean, it's essentially indiscernible from a live action video, but Byte Dance is huge and they invest a lot of money into this. And so my sense is like, that's a very, very, very expensive proposition. 1:19:58 Speaker 1 Yeah, understood. Well, Eric, I really enjoyed having you on this podcast today. 1:20:04 Where to Find Eric Where can people find out more about you? 1:20:07 Speaker 2 Yeah, I'm on X probably too much, but you could go to mdm.co, Mobile Dev Memo, mdm.co and yeah, I, I, I blog there, run the blog, run the podcast, run the newsletter and have done for some time. 1:20:24 Speaker 1 Yep and highly recommend his blog for all things at Tech and his podcast is great as well too. Thank you again and until next time. 1:20:32 Speaker 2 Yep, cheers.