Performance Marketing as Arbitrage

The definition

Adam Foroughi gives an operating definition of performance marketing that most people, in his account, do not use. Real performance marketing, he argues, makes the advertiser certain that the dollars they put in return more than they take out. When that certainty holds, the advertiser becomes an arbitrager of the platform, and in his words "the only constraint on them scaling in our system is the money that they have in their bank account."¹ The advertiser plugs in, spends, and the platform's measurement tells them, with confidence, that they made more back than they spent. A rational advertiser then puts in as much as they have, and the platform's job reduces to driving as much scale as possible before the arbitrage breaks down, meaning before incremental spend stops clearing the advertiser's return-on-ad-spend goal.

Foroughi frames the alignment as total: "more value is get the advertiser wealthier. They get wealthier, we get wealthier." Because the return is certain and measurable, there is no fixed budget to fight over and no salesperson required. He contrasts this with the brand model that AdMob was built for, where ad performance is "completely hand-wavy" and "the biggest cut goes to the folks that are wining and dining the client the most." That, in his description, is a sales business. He deliberately refused brand advertisers to stay "in the business of no sales, high-value product," where the advertiser's own profit and loss statement, not a sales pitch, proves the product works.

The model is the business model

In Foroughi's telling the accuracy of the recommendation model is what makes the advertiser's return certain, which makes the moat and the business model the same thing. He describes AppLovin's Axon as "not a large language model, so there's no consumer interface," judged on a single question with every iteration: can it drive more spend at the advertiser's return goal, with more reach. He points to the 2023 jump from Axon 1, which required heavy spend and manual labor to reach the arbitrage, to Axon 2, deep-learning-based, where any advertiser can plug in and immediately arbitrage. That shift, in his account, is what made the loop scalable enough to inflect the whole business.

He roots the origin of the model in a structural advantage of gaming. Because gaming revenue flows through Apple and Google, gaming advertisers could close the attribution loop precisely, spending a known amount on an install and knowing what it would generate over time, with 2019 install costs ranging from fifteen to thirty cents for hyper-casual titles up to fifty to a hundred dollars or more for core games in the Game of War tier.² Cross-platform businesses, with revenue spread across web, app, and physical channels, could not close that loop and carried a permanent modeling disadvantage. This is why, on his account, gaming companies rather than direct-to-consumer brands became the dominant mobile advertisers: they could model the arbitrage, so they scaled spend to their bank balance.

The structural properties Foroughi claims

Foroughi argues the model is category-portable, since the same loop carried AppLovin from game ads toward e-commerce and, in principle, any small business: the audience and the measurement stay the same, only the advertiser changes. He also claims structural fraud resistance. Because the platform sells installs and post-install activity rather than impressions or clicks, a fraudster would have to spend real money to generate real behavior to profit, and "fraudsters are not willing to spend money to generate fraud then they lose money." Foroughi's instinct throughout is to make the technical accuracy of the model, rather than a sales organization, the thing that grows revenue.

He is candid about where the frame is exposed. The whole loop depends on trustworthy measurement: if attribution is wrong or gameable, the arbitrage frame collapses into the same hand-waving he criticizes in brand advertising. "Arbitrage until it breaks down" also implies a natural ceiling per advertiser, so platform growth eventually requires more advertisers, more inventory, or a more accurate model that pushes the breakdown point out. And whether the certainty that held in gaming extends to arbitrary small-business categories is, by his own framing, the unproven bet. The advertiser's side of the same asymptote, taking a channel to saturation until the arbitrage stops clearing, is described in saturate the winning channel. The consumption logic that a single measurable driver dictates the whole model appears again in purchase volume as the single variable. ²