Season 4, Episode 11: Understanding AppLovin

The sponsor of this week's podcast is Clarissites. Do you juggle multiple tabs and tools just to understand your marketing performance? Are you tired of slow BI dashboards and manually aggregating data into spreadsheets? What if you could access all the data you need in one place? Without dependencies on SQL, spreadsheets, or valuable data team resourcing? That's why the performance marketing teams at Uber, HelloFresh, DeliveryHero, and many others trust Clarissites to run marketing reporting, automating all manual work and helping them focus on insights that drive performance. What used to take weeks now takes minutes with Clarissites. Go to Clarissites.com/demo to check it out yourself. You'll see why thousands of performance marketers trust Clarissites every day. The problem is that the distinction needs to be drawn between the confidence of the economists and the correctness of their analysis. Welcome to the Mobile Dev Memo podcast. I am your host, Eric Sufra. This episode of the podcast is in a different format from what I usually publish, and then I don't have a guest. Instead, I'm going to present a long-form analysis as a monologue. I experimented with this format back in season 3, with the DMA personalized advertising and digital decalobilization, and it was well received. So, I'm doing it again. The topic of this podcast is apploven. I'd imagine that most listeners are probably familiar with apploven, and the dramatic rise in its stock price over the past few days. The title of this podcast episode is "Understanding Apploven," because I want to try to elucidate the company, decipher its business model, and help a broader audience understand its astonishing growth. I've been a customer of apploven in my roles as a full-time operator, and now, part-time consultant since his far back is 2014. And I've covered apploven on Mobile Dev Memo since 2015. Apploven is no-pate company, and it's clear why many, but especially public markets investors, are bewildered by its meteoric recent growth. And that growth has been meteoric. As I write this on November 9, the company's stock price is $292 for a market cap of $97.3 billion. One year ago, on November 9, 2023, Apploven's stock price closed at $39.70, and the company's market cap was $13.4 billion. The company's stock price has grown faster than Metas. As I write this year to date, Apploven's stock is up by 557% versus 71% for Meta. And while much of the stock price movement is recent, following the company's earnings report last week, the average stock price in October 2024 of $150 is up by 288% from Apploven's average price of $39 in October 2023. It is certainly possible that Apploven has attained meme stock status and that some of the recent price increase will be given up. And while I've seen people claim in the last week that they were certain that Apploven would ultimately become a $100 billion company, that didn't necessarily seem obvious from the outside. In fact, Apploven was set to be acquired by Chinese private equity firm in 2016 for $1.4 billion, but the deal was scuppered by the committee on foreign investment in the United States or CIFIUS. And in fact, Apploven's revenue growth accelerated fairly recently, following the shakeup of the mobile gaming advertising market, catalyzed by Apple's app tracking transparency privacy policy. And that revenue growth acceleration has indisputably been considerable. Apploven generated $504 million with its advertising platform in Q3 2023, growing to $835 million in Q3 2024. And in that same period, adjusted EBITDA for the advertising platform grew from 72% to 78%. I offer my speculation, and I am just speculating, for how the company achieved this growth in this episode of the podcast. But understanding Apploven the company is fundamentally predicated on understanding the market in which it operates, mobile games advertising, and the direct response digital advertising landscape more broadly. And so this podcast is structured in three parts. First, understanding direct response advertising, second, understanding mobile gaming advertising, and third, understanding Apploven's growth. All of these topics can seem unapproachable and impenetrable from the outside. I hope this episode clarifies them. The commercial model used in direct response advertising is particularly misunderstood. I presented a framework for thinking about the distinction between direct response, what I call delayed response, and brand advertising in a piece I wrote in 2021 called the Parallelist Mythology of Brand Marketing for Digital Products. This is how I framed the concept of direct response advertising in that piece, and I'm quoting. Direct response marketing is almost exclusively accomplished through digital advertising for digital products. And its purpose is to foment an impulse on the consumers part to immediately purchase or otherwise engage with the product. Ideally, this opportunity is guided through audience and intent targeting, such that the advertisement collides with need or desire at an appropriate time and results in an outcome for the advertiser, known in advertising parlance as a conversion. End quote. I described direct response advertising and brand advertising and delayed response advertising as tactics in that piece, which all fall under the broader umbrella of a strategy called performance marketing. Performance marketing as a practice simply captures the notion that all marketing efforts are pursued under the auspices of some quantitative model, such that each dollar invested returns more than a dollar. To my mind, there is no alternative approach to performance marketing, or at least there shouldn't be. I certainly hope that no marketing teams are spending money that they credibly believe will not be recovered. When a marketing team invests in direct response advertising, it buys a piece of advertising inventory with the expectation that the person who sees it will immediately take some commercially relevant action. The use of the word expectation here is deliberate. Obviously, not every ad impression leads to an outcome for the advertiser. In fact, the overwhelming majority do not. But direct response advertising under the performance marketing model results in average outcomes, in statistical terms, the expectation of returning more money than was spent. Given the skewed nature of advertising outcomes, this means that some small minority of users react to ads in ways that contribute vastly more revenue to the company than it spent in reaching them, compensating for everyone else. This is a powerful idea. If a company can credibly, guided by analytics, spend one dollar to return more than a dollar nearly instantaneously, then it can essentially buy revenue. And with a number of caveats, this is precisely how direct response advertising, in success, performs. One of those caveats is that it is very challenging to reliably predict the average value of advertising on some advertising channel. In the marketing vernacular, the return on investment of advertising expenditure is return on ad spend, or ROAS. ROAS can be challenging to measure, let alone predict. It requires knowing how a consumer was introduced to a product through a process called attribution, which has become less clear over time as various ecosystem privacy restrictions have made identity data less available. But beyond attribution, a marketing team must also predict not just the revenue that will be directly generated from a consumer's interaction with an ad, but all subsequent revenues within some amount of time following that interaction. This idea is known as lifetime customer value, and it is the subject of a large volume of academic literature. Essentially, an advertiser wants to ascribe as much value to an ad as is reasonable to allow the advertiser to spend as much to reach potential customers as possible. Imagine a person that sees an ad for pants from some retailer. The person clicks on the ad, buys the pants, then one week later buys a pair of shoes from that same retailer. And one week after that, the person buys a shirt from that retailer. How many of those purchases should the advertiser consider to be the result of the initial ad exposure? On average, what time period following an ad exposure should an advertiser consider in evaluating the performance of that ad? And what if the ad wasn't clicked on at all? But merely viewed. These are the considerations that make lifetime customer value thorny. But it's vital to do well for direct response advertisers because it informs how much they can bid on advertising inventory. For the most part, digital advertising is mediated by an auction system. Whenever an ad impression becomes available within an app or on a website, an auction is run in which advertisers bid for that impression. The advertiser determines how much that particular piece of inventory is worth to it, or rather uses an algorithm to price inventory and submits its bid, with the highest bidder winning the inventory and oftentimes paying a price that is slightly more than the second highest bid. The advertiser captures whatever the difference is between what they pay for the inventory and what they believe the inventory is worth to them and profit. So while the row-as of the inventory is the advertiser's projected revenue from buying the impression, divided by the cost of fillet, its profit is the difference between the value gleaned from filling the inventory and the cost of fillet. Because of the nature of calculated customer lifetime values, advertisers often time index the row-as metric by citing, for instance, a 90-day row-as or a 120-day row-as. This describes whatever return the advertiser expects to generate, revenue divided by cost within that period. A 110% 90-day row-as means the advertiser's expectation or its average projected yield on $1 in advertising expenditure is $1.10 over 90 days. Sometimes this is expressed as p-row-as or predicted row-as given that it's evaluated with the prediction of revenue. Row-as might also be stated as i-row-as or incremental row-as, meaning only the purchases that are deemed to be incrementally generated by advertising. In other words, the purchases that are deemed to have only been the result of average revenue. advertising, and which wouldn't have happened without it, are considered in the equation. Whichever row-ass flavor is used, it guides the direct response advertising apparatus utterly. An advertiser will establish some row-ass standard within a timeline, and this is usually determined by the company's CFO, and spend as much money as a can on every channel that can deliver that level of row-ass. This concept can be counterintuitive and confounding. The direct response marketers often don't face budget constraints, but rather hard row-ass constraints, and they seek to maximize advertising spend on any channel that can support those row-ass constraints. In other words, a marketing budget isn't a fixed pie that gets allocated to advertising channels based on their performance, but rather the fluid outcome of being able to achieve row-ass goals wherever possible. The reason that budgets aren't generally constrained is that the payback periods for digital advertising can be short enough to nearly endlessly finance. Consider a 90-day row-ass of 110%, which isn't unrealistic for many digital products or e-commerce shops, and in fact, may be conservative. What investment opportunities offer 10% yield in 90 days? Accounting for payment periods, which are often 30 days, an advertiser could support growing spend against 110% 90-day row-ass through short-term loans or even credit cards. Why would overall budget be a constraint if money is recovered so quickly? It's important to consider here that that money isn't just recovered within 90 days, it's redeployed in compounds within 90 days. The revenue generated by advertising spend is delivered through some series of cash flows that, on average, if the advertiser's models are robust, will have delivered the desired return of 110%, 90 days after the money is deployed in the 90-day row-ass example. But this isn't like a bond that returns principle on some maturity date. The money that an advertiser spends on advertising is recovered, in part, every day. In fact, the advertiser will have recovered some percentage of its advertising spend on the same day the money is deployed. While there are details for when that revenue is paid out that need to be considered, the advertiser can recognize it and effectively redeploy it, along with revenue being contributed on that day from all previous cohorts. This compounding effect of short-term advertising investments presents a profound commercial opportunity for the kinds of companies that can effectively implement direct response advertising. By my estimation, those conditions are minimal distribution costs and a large total addressable market. Mobile gaming is perhaps the single canonical product category, optimally suited for benefiting from the dynamics of direct response advertising. The sponsor of this week's episode is incremental. Were you looking for an alternative to attribution and went down the route of MMM, but after a four-month integration the results were confusing and unhelpful? Have you thought of incrementality measurement? Incremental is the first and only always on incrementality measurement platform that doesn't rely on user-level data, planned experiments, or MMM. The platform is trusted by some of the top mobile marketers. Check out the case studies section on their website to learn why. Join the future of marketing measurement with incremental. Google "Incremental" the future of measurement or use the link in this episode's description to learn more about why it should form part of your marketing tech stack. I spent the first portion of my career in mobile gaming and wrote a book, "Freemium Economics," dedicated to the freemium business model. I can make the case that mobile gaming satisfies both of the conditions for benefiting from direct response advertising. There are no distribution costs for mobile games. They are distributed principally in platform app stores, Google Play and the iOS app store, which hosts them for free and take a cut of their gross revenue. Once a game has been developed it costs nothing to a developer to distribute each copy, and, critically, since the overwhelming majority of mobile games and essentially all commercially successful mobile games operate under the freemium model, it also costs consumers nothing to play a developer's game. Developers make money from their games through the sale of virtual in-app products, often denoted as IAPs, and from selling advertising inventory within their games. The majority of in-game advertising inventory is purchased by other games, although that share is decreasing, which I'll get to later. Advertising is the principal engine of user base and thus revenue growth for most commercially successful mobile games. Obviously exceptions exist, but outside of the handful of household name titles that benefit from entrenched virality. Revenue for mobile games is a function of advertising expenditure, which is termed as user acquisition. Mobile game developers utilize several different types of advertising channels for user acquisition, but the primary destinations for mobile gaming ads are social media and other games. The idea that a mobile game developer would sell advertising inventory to other mobile game developers, including its direct competitors, might sound counterintuitive. But the market dynamics of this are generally sound as a quirk of the freemium model. Only a very small percentage of the players in mobile games purchase in-game items, almost universally fewer than 5% do, and sometimes fewer than 2%. Given these low purchase rates or conversion rates, advertising is the only means of deriving economic value from the vast majority of mobile gamers. Developers will often limit the exposure of ads to only players that are deemed unlikely to ever make a purchase in the game, so that payers and potential payers aren't unduly funneled into a competitor's game. There are two important and somewhat counterintuitive concepts to unpack in this idea of a market for mobile gaming inventory amongst mobile games. First, one might expect this to be a market for lemons, given that only non-payers are seeing ads. There's obviously asymmetric information in this market. The seller knows everything that the player has done in their game, and the buyer knows nothing about that. But the fact that a game is willing to show ads to a user, to sell that user, provides critical information to the buyer that the user is unlikely to have monetized in that game. All buyers know this, since most game developers are both buyers and sellers of inventory, and limit ads to non-payers. So while total information about users is asymmetric, the notion of intent to purchase isn't. And in any case, it doesn't really matter. There are myriad reasons why a potential payer wouldn't pay in some specific game. Buyers are simply being optimistic, presumably with support from their historical ability to convert players. The second concept is that showing an ad for a game, with the expectation that the user might click on that ad, download that game, and then not return to the original game, is a better way of monetizing users with ads than through a succession of brand ads over time. This might strike some as odd. The successful outcome with an in-game ad for another game is that the user adopts the game being advertised, which probably involves leaving the game that showed the ad. Why not just show a series of TV-like ads in the game over time, which would result in the user staying in the game? There is a sensible and eminently rational reason to not do that. When gaming advertisers price their bids, they do so on the basis of the price of various types of conversions, but often installs. That is to say, a mobile gaming advertiser bids on installs not impressions. They expect the advertising channel to deliver an install, and their bid represents the price they are willing to pay if that install is delivered. Again, that bid value should be lower than the advertiser's expectation of revenues. The delta is the advertiser's profit. And these bids, often called cost per install targets, are almost universally higher than what any brand advertiser could bid for the inventory, and they're generally higher than what any other digital product developer could bid for any other conversion outcome, including an install. So the reason a developer sells its inventory to other games isn't because it explicitly chooses to. Keep in mind that the inventory is sold in an auction process. The reason that games sell their inventory to other games is simply that games win those inventory auctions. This is a critically consequential idea to comprehend in understanding digital advertising generally, but especially the mobile gaming advertising market. It's helpful to explore it from the other side, from the perspective of an advertising channel. Here I'll talk about advertising platforms and advertising networks. I use these phrases in specific ways. In my mind, an advertising platform sells its owned inventory, like a social media platform does. Think of advertising platform as being synonymous with walled garden. An advertising network doesn't necessarily own inventory, but it connects buyers, advertisers, with sellers, publishers. It helps advertisers buy the most relevant and valuable inventory by applying pricing logic informed from the data it glends, from operating across all of its advertising clients. Ad platforms and ad networks essentially offer advertisers abstractions. Tell us what you are willing to pay and we'll find inventory for you at that price. But modern ad platforms and networks don't traffic inventory. They don't sell impressions directly. What they sell are outcomes or conversions. They do this through a process called conversion optimization, which I unpack in detail in a piece I wrote in August 2023 titled Understanding Conversion Optimization and Digital Advertising. With conversion optimization, an advertiser sets a price for some outcome, say a purchase or a registration or an ad to cart action on an e-commerce shop. And the ad platform runs various tests with the advertiser's ads to see whether it can deliver those conversions for the price the advertiser wants with some margin for itself. Ad networks and ad platforms take all the risk in this configuration. If they fill an impression with an advertiser's ad and that ad impression doesn't lead to the outcome, the advertiser bid against, then the advertiser owes nothing, and the ad platform or network wasted that impression. There's opportunity cost here. If a different ad from a different advertiser would have produced a conversion, that revenue is foregone. So ad platforms and ad networks are incentivized to determine what the probability is of an ad leading to a conversion and to wait the bids they're sent by advertisers by the likelihood that their ads will result in the outcome that is being bid against. In this way, the ad platform or ad network is also calculating an expected value of filling the impression. The bid times the probability that the impression would lead to the desired conversion. This is how ad networks and ad platforms rank ads. Two advertisers could submit wildly different bids, but if their conversion probabilities are also substantially different, it's possible that the lower bid produces a higher expected value and wins the auction. All-purpose ad platforms like Metas and TikToks and Google's can service a number of commercial outcomes because their client base is varied, as is the content that hosts the inventory they are selling. Mobile gaming ad networks tend to optimize to installs with costs per install bidding, but many can also optimize to other outcomes like retention. In this case, an advertiser might bid against a user retaining or being present in the game for some number of days. There is another class of advertising tools here that is important to mention, but that I won't go too deep into the weeds on. These tools are demand side platforms or DSPs and supply side platforms or SSPs. DSPs give advertisers direct access to inventory that is sold on exchanges and SSPs handle yield management for publishers. These tools operate in what's known as the programmatic market because they facilitate real-time bidding on inventory. DSPs are often offered as as-is tools that require advertisers to provide their own specific targeting and purchasing logic, although the lines between demand side platforms and ad networks have blurred considerably within the mobile gaming space. And the role that SSPs play in the mobile gaming advertising market has changed systemically in the past few years as a result of Apple's app tracking transparency or ATT privacy policy, which reached a majority of iOS devices in July 2021. I've written extensively about ATT on mobile dev memo, so I'll just provide brief background here. For anyone interested in better understanding the details and consequences of ATT, mobile dev memo is a good place to start. Prior to ATT, mobile gaming attribution on iOS was managed through a device identifier called the IDFA or ID4 advertisers. This identifier is unique to iOS devices. Because of a historical peculiarity in the way meta allowed its own inventory to be attributed, as well as the mechanics of sending traffic to app stores, mobile gaming advertisers work with third-party attribution companies called MMPs that intercept ad clicks, record the device identifiers of the users that clicked on them, and then forward those users onto the relevant destination in the app store or Google Play or wherever else. Then, when those users downloaded the apps they clicked ads for, the MMPs would observe those device identifiers in the apps, either through an SDK or what's known as a server-to-server postback, and attempt to match them with recent clicks. When those matches were observed, those attributions were credited to the ad networks that delivered those installs. Note that I'm only describing the process for mobile gaming advertising here. The attribution processes for other advertising categories, especially web-centric categories like e-commerce, were different. For mobile gaming, the attribution process didn't stop at the install. Because advertisers in ad networks and platforms alike want to understand the total value of installs, the app would post back other events like purchases to its MMP, which would tally those events against the install. In this way, advertisers could keep track of total revenue per install or RPI, and ad networks could understand just how valuable a given user was for a particular advertiser. Ad networks were able to create ledgers of these IDFA's and their app spending over time. This was useful for targeting, an ad network or platform could specifically target ads to IDFA's that were known spenders in order to increase the likelihood that a given ad delivered conversion outcomes. But ATT obfuscates the IDFA for the majority of app users on iOS, which, for the most part, although high-profile exceptions do exist, is the most important platform for mobile games publishers from a revenue perspective. I'm glossing over very many details of ATT here in the interest of time, so I'll recommend that anyone who isn't familiar with ATT read the blog. Mobile game economies are driven by a very small minority of high value players, the people who spent thousands, tens of thousands, or even hundreds of thousands of dollars in a game. Because of this, it was important for mobile gaming ad networks to understand which users provided revenue to the games they installed. The obvious use case for this is targeting. Ad networks would want to target historical spenders in ad campaigns. This is true for large, scaled ad platforms like Metas, but I'd argue that it wasn't the principle benefit of the IDFA for mobile gaming ad networks. Outside of very specific categories like social casino and certain forex games, I don't think mobile gaming ad networks utilized IDFA lists for direct targeting to great effect. There are a few reasons for this. First, conversions are very rare. There simply isn't enough of this data to inform a targeting strategy broadly. More inputs are needed in order to build a traffic acquisition strategy for a mobile game developer. Second, these users are highly coveted and everyone knew who they were. There is no alpha in targeting a high value player when everyone knows they are a high value player. The price will converge around the true value. And third, demonstrably engaged players, that is, the ones that advertisers would like to reach, are difficult to poach precisely because, totalologically, they are playing the games they are monetizing in, and they may be impossible to reach in that game given the advertising logic I described earlier. Unlike Metas total visibility of consumers across the websites and apps they use, mobile gaming ad networks only observe users in games and only in the games they are unlikely to monetize in. This is a very limited view. The IDFA was certainly used as an input to targeting decisions by mobile gaming ad networks, but I'd argue that it wasn't the principal input simply because it couldn't be used that way. Mobile gaming ad networks used the natural contextual relevance of games for targeting. The fact that one game was similar to another game. Again, IDFA indexed data was certainly layered into targeting decisions, but by and large contextual relevance was likely to be a more meaningful input to targeting. The IDFA was useful for aggregating data at the cohort level to understand how well that contextual targeting worked at delivering outcomes. The IDFA in this use case isn't applied to building behavioral histories of individuals for targeting, but rather for understanding whether advertising game A and game B tended to produce more average revenue per user than the advertiser was paying in CPI. In other words, the IDFA was used to get a sense of the average value of a user in one game when they were introduced to it by an ad from another game. This idea will resurface in a few minutes. Absent the IDFA, mobile gaming advertiser couldn't really do this. Mobile advertising attribution tools reverted to IP address-based fingerprinting with the IDFA's disappearance, which can be used for install attribution, but is unreliable for purchase attribution when device IP addresses change over time. So what mobile gaming advertising networks lost with ATT was, yes, the ability to understand individual gaming behaviors over time. But much more importantly, to understand when campaigns were working to deliver value so that those campaigns could be tuned to advertiser expectations. This week's episode of the mobile dev memo podcast is brought to you by Vipe. Powered by proprietary data and machine learning, Vipe is the leading streaming TV ad platform for small and medium-sized businesses looking for actionable campaign performance. Vipe's newest suite of AI-nabled products like Vipe Studio, Vipe AI Assistant, and Vipe IQ2 are just the latest in a string of intuitive and transparent tools designed to radically democratize access to premium television advertising. Try them out for yourself at vibe.co/sina. That's vibe.co/sina. MoPub was an in-app SSP. MoPub was acquired by Twitter in 2013 for $350 million. AppLeaven acquired MoPub from Twitter and this is pre-elawntwitter for just over $1 billion in October 2021, roughly three months after ATT reached majority scale on iOS devices. I wrote about the acquisition at the time in a piece titled "Why Did AppLeaven By MoPub?" I'm quoting from that piece but not continuously as I'm jumping around. Quote. I think the answer is mostly obvious and aligned with AppLeaven's general operating strategy, at least since it acquired Macs, its in-app header bidding product. To aggregate as much mobile advertising supply as possible such that it can construct a corpus of data that provides a competitive advantage to not only its advertising platform but also to its first-party content business. AppLeaven's acquisition of MoPub contributes to this strategy by giving AppLeaven SDK access to MoPub's existing publisher client base by eliminating a major supply competitor and by delivering an advantage to its demand platform app discovery. Supply platforms get insight into every impression filled on behalf of publisher clients, bid levels, fill rates, etc. Through an in-app bidding platform such as Macs, AppLeaven receives this data in real time and can aggregate it in ways that benefit all aspects of its business, which is especially valuable in the post-ATTE environment. Broker ad networks like Apploven, and Unity, Bungle, IronSource, etc. Almost exclusively service in-game inventory, and aren't as impacted by the deprecation of the IDFA as is Facebook, because they had access to and collected much less user-level data. Games advertising is primarily driven by contextual targeting, ads for GAMEA perform well when they are placed in GAMEB. This narrowly-scoped data is obviously more valuable in higher volume. This data is helpful to the supply side of the business, because it provides transparency into which apps are bidding on what inventory, at what values, and through which networks. It's helpful to the demand side of the business for roughly the same reason, but from a different angle. It helps the demand platform to understand which publishers are most competitive, and from which advertisers, allowing the demand platform to price bids accordingly. It's beneficial to the bidding platform, MAX, because that data can be used as leverage to incentivize its adoption. And MOPA may have been the last real supply provider available for acquisition, given that almost all other mobile advertising platforms have both the demand and supply component, and most are building bidding or mediation solutions. Okay, quote over. That was a lot, but it essentially captures the idea from earlier that mobile gaming advertising networks need to be able to derive an expectation for advertiser outcomes. If they can't do that from post-installed data, they can approximate it by looking at what all advertisers are bidding for all inventory. In essence, they rely on an advertiser's ability to calculate a useful customer lifetime value, to conduct price discovery, to understand what certain inventory is worth, under a number of different dimensions, such as the app bidding and the app selling. This data is available from the SSP, since the SSP receives all the bids for the inventory, and that data can be used to the benefit of crafting bids on behalf of advertisers. As I noted in the piece, the SSP market had already consolidated to just a small handful of players by the time AppLove and acquired MOPAB. In fact, AppLove and already owned an SSP, Max, when it acquired MOPAB. It shuttered MOPAB and paid $210 million in publisher bonuses in Q1/2022 to transition MOPAB clients to Max. My understanding is that level play, an SSP offered by IronSource, was the second largest in terms of customer footprint, prior to AppLove and's acquisition of MOPAB. In acquiring MOPAB, AppLove and Leapfrog'd IronSource, with its Max product becoming the largest mobile gaming SSP. And Max is almost certainly the largest by a comfortable margin, in the company's Q4/2023 earnings call, after which it stock jumped by 30%. AppLove and revealed that two-thirds of the top downloaded games are clients of Max. IronSource is a mobile gaming advertising network that merged with Unity in November 2022, although the merger was announced in July 2022. I hypothesized at the time of the announcement that the merger was being pursued for exactly the same reason, because Unity, which builds Unity 3D, the most popular mobile game engine, and operates a mobile advertising network, needed access to an SSP for its data. Unity was a laggard to the SSP space, and only announced its own entry into the SSP market, the same month that AppLove and announced it would acquire MOPAB. To add to this drama, AppLove and pursued a merger with Unity in August 2022, the month after the IronSource merger was announced, although it abandoned that bid. This race to capture SSP data indicates how valuable that data became in the aftermath of ATT. I wrote about this in July 2022, the same month that Unity announced its merger with IronSource, and a piece titled, "Why Mediation is the Primary Front in the Mobile Advertising Wars" post ATT. I'll quote from that piece, and again, I'm jumping around. The reason for this is that aggregation of supply provides pricing insights that are critical for efficiently managing demand in the post ATT environment, in which user-level data cannot be gleaned and aggregated, either into user profiles or simple device graphs. If targeting can't be accomplished and optimized by understanding how individual users behave after being exposed to ads, then ad platforms must acquire that awareness using the next best thing, Advertiser Bid. Note that an advertiser's bid for inventory is an effectively in approximation of the LTV that the advertiser expects that prospective user to generate in its app, adjusted by various conversion filters, click rate, install rate, etc. In essence, a bid is an LTV estimate, and SSP can interpret a bid as the pairwise expected value of a user in AppA, the advertiser's game, when acquired from AppB, the publisher's game. Broker ad networks don't necessarily need data at the level of an individual user, since they primarily target contextually, that is, app to app, they really only need to dimensionize data across various features, like phone model, geography, etc. But for the most part, the limits of SK ad network preclude even that. And so app-lovings acquisition of MoPub and Unity's merger with IronSource makes sense for this reason. They're both acquiring data from SSP operations that can be used to benefit advertiser campaigns with demand-side operations, end quote. If this is all true, then it partly explains AppLumman's growth. AppLumman acquired access to a fire hose of data that it can use for price discovery in a data environment that has been systematically stifled and constrained by platform privacy policy. I say partly because I don't think it's possible to overstate the degree to which Unity's pace and quality of execution has been hindered since ATT was rolled out. Unity's merger with IronSource and the engine runtime fee policy introduced later that led to the departure of nearly the entirety of its management team created operational friction for the company. I made the case earlier in this podcast that ad networks don't necessarily compete with each other in the traditional sense of the term because direct response advertising budgets are not zero sum. The mobile gaming advertising market can be grown in mutually reinforcing ways when participants deliver performance to advertisers. And while that's true, SSP usage is mutually exclusive and the distraction of the cross-border merger between IronSource and Unity as well as the replacement of the company's entire C-suite surely erected some barriers to level play adoption. One aspect of the expansion of AppLumman's SSP footprint through its MOPAB acquisition and thus its access to SSP data that I think is underappreciated is how it likely disproportionately creates margin opportunity with smaller games as a measure of DAU than larger games. I discussed this idea on my recent appearance on the Shriteri podcast last month but it's worth reiterating. Ad networks make money from demand by filling impressions at one price and charging their advertising clients another, which is their bid for an outcome. Smaller games are less understood in terms of player value than larger ones precisely as a function of scale. Large games sell large volumes of inventory and advertisers and ad networks alike can flesh out distributions of value across many dimensions. Smaller games by definition don't do that. Prior to ATT, advertisers might have bid on specific IDFA's known to belong to high-value players through programmatic channels but that inventory was mostly uncompetitive for ad networks with contextual targeting. But with SSP data, an ad network can understand how those smaller games themselves bid for users, which is a proxy for value. And because without SSP data, the inventory of those games remains uncompetitive and ad network with SSP data can likely capture very high margin on that uncompetitive inventory. This data is valuable in and of itself but AppLuv enrolled out a major upgrade to its advertising targeting and optimization engine Axon in Q1 2023 after its acquisition of MOCUP. Details are light on what that upgrade entailed. My sense is that it absorbed many small component level targeting models into one larger end-to-end model with an expanded feature space. This approach would be similar to that taken by Meta for content classification with Reels and it would be corroborated by the fact that AppLuv enclaimed to have, quote, "one of the largest GPU deployments in the world at this point," end quote, in the company's most recent earnings call. It bears repeating that the hypotheses presented in this podcast are speculative. And while they may explain the growth in AppLuv and revenue, which is a determinant in the company's stock price, they don't explain the precipitous rise in the company's stock price, particularly after its most recent earnings call. That, to my mind, is better explained by AppLuv and expansion into e-commerce advertising. Apparently, in its pilot state, this initiative is going well. The e-commerce advertising market is larger than that of mobile games advertising. e-commerce is and has consistently been met as largest advertising category by revenue, for instance. If AppLuv and Axon targeting an optimization engine is indeed able to efficiently price in-game inventory for e-commerce advertising, it certainly represents a meaningful growth opportunity for the company. And yet, mobile gaming remains important. And so I hope this overview of the mobile gaming advertising market, with a broader explanation of the direct response advertising model and mindset, has contributed something helpful to the corpus of information available on both subjects. And I also hope that it has helped listeners in understanding AppLuv. [MUSIC]