Over the past few years, we’ve built the most advanced AI-driven advertising platform in the industry — powered by a lean, exceptional team. Since launching Axon 2 in 2Q23, advertising spend on our platform has roughly quadrupled, making AppLovin the highest-valued advertising company in the space. Our rapid rise reflects the strength of our technology and its impact, but we haven’t always paused to explain how we do it. This blog dives into our business model, its merits, and our approach to data — offering a clear view for our investors, partners, and anyone curious about what drives us.
Our business model: creating growth, not just ads
At AppLovin, our mission is straightforward: We drive incremental revenue for advertisers. Every product we build ensures that when advertisers invest with us, they see measurable returns — growth that outpaces their spend. This keeps them thriving, and it’s how we’ve scaled without a large go-to-market team. Let’s break this down.
Mobile gaming: fueling discovery and expansion
Mobile game developers craft incredible experiences, entertaining billions monthly. But in a crowded market, organic discovery is tough. Advertising bridges that gap. When it works, the market grows; when it falters, growth stalls. The Western mobile gaming market boomed from 2012 to 2021, then hit a wall in 2022 — often blamed on post-COVID, though post-IDFA marketing challenges were the deeper issue. Axon 2 was one of the main market catalysts that turned it around. Since its launch, we’ve sparked a resurgence: In-app purchase (IAP) revenue is growing mid-single digits annually, while our MAX publishers are growing several times faster. How? We’ve scaled ad spend for gaming clients to roughly a $10 billion run rate annually — up around 4x in the two years since Axon 2 — unlocking discovery and revenue that powers the ecosystem. Had we not created this innovative, breakthrough technology, the industry would still be struggling today.
Web advertising: unlocking new channels
For e-Commerce and other web businesses, discovery is just as critical. Many have leaned heavily on Meta, but reliance on a single channel caps growth and squeezes margins. We’ve created a new opportunity — not one that diverts spend from other channels, but one that gives merchants a fresh space to spend in and should create expansion of their businesses. We’re not perfect yet; our web product is on an early-generation ROAS model, alignment with external tools is evolving, creative designs are limited, and our self-service and agency dashboard hasn’t been released. It took us nearly a decade to hit a $1 billion annual spend run rate in gaming; on the web, we reached that in months. Full integration with third-party platforms and improved optimization are on the way.
How Axon works:
Technology, not tricks
What fuels this growth? Axon, our AI engine, delivers real revenue results for advertisers. Think of it like the LLMs behind tools like Grok 3 — built from scratch through brilliant engineering, not shortcuts. There’s no hidden data trove. Axon taps five buckets: MAX loss notifications (standard, commoditized data all bidders get), advertiser data, gaming usage patterns, third-party data from mobile SDKs and web pixels, and user engagement data from our ads. The magic lies in our models’ sophistication, supercharged by a reinforcement loop. Serve an ad — like one with a mini-game — and we get dozens of interactions back. That data sharpens our predictions, building a moat: The more we serve, the smarter we get. Scale fast, learn fast — it’s AI’s winning formula, and we’ve mastered it.
How it works in our world
Data powers modern advertising, but it’s also where questions arise. Let’s unpack how we handle it across apps and websites.
Apple’s App Tracking Transparency (ATT) reshaped in-app advertising. Users can choose to share IDFA for cross-app tracking — or not. When they opt out, we don’t create alternative accurate and persistent identifiers, typically called device fingerprints.
So, how do we keep ads relevant? Our models evaluate a wide range of signals — some specific (when users consent), some general — to statistically estimate which ads are most likely to drive engagement or conversions at that moment.
For example, if a new user opens an app for the first time, the model might consider signals like app context, recent ad performance, or the user’s IP range (which can briefly reflect general location or shared browsing behavior). These signals are ephemeral and non-identifying but are still useful in a cold-start scenario. As more interactions occur, the model learns and adapts — without ever needing to persistently identify the user.
In this environment, IDFA remains extremely valuable — it allows the system to connect engagement over a longer time window. But it’s not essential. On MAX, US full-screen ad CPMs are roughly double with IDFA versus without — a clear indicator of the market value of that signal.
The data we don’t touch
A few firm lines: We don’t buy or sell data from brokers — our info comes straight from partners who choose to share it with us solely in the context of providing them with advertising services. It also comes from our own tools, with no emails, phone numbers, or any other data that could be used to triangulate a person’s real-world identity.
In apps, we comply with ATT on iOS. Our SDK collects only basic device information from public APIs the OS provides and allows, which is consistent with every major SDK. We don’t tap Adjust data beyond what advertisers explicitly choose to share with us; it runs on separate infrastructure, unchanged since the acquisition, with attribution logic fully independent of our influence. With MAX, we use only standard win/loss notifications every bidder gets; the bid stream data stays separate and is purged after seven days.
The open web, built on decades of cookies and pixels, operates differently and isn’t bound by ATT. Advertisers embed our pixel, which feeds our models with audience behavior to optimize ad delivery.
On Crocs.com, for instance, check the pixel there: No third-party cookies or IDs are appended, because we neither need nor request them. Contrast that with TheWoobles.com, where you might spot extra IDs (Facebook’s, Snapchat’s) on our pixel. That’s not us — Elevar, a tool used by the advertiser for their own analytics, appended them to our pixel. We don’t ask for or use this data; it’s purged when it hits our servers. Or take TrueClassicTees.com: our pixel gets tagged with an “igId” from Intelligems.io, an A/B testing platform the advertiser employs — not an Instagram ID, despite appearances. Again, we don’t use it. Our models rely solely on what we request and expect to receive — advertisers can’t overload us with extras.
Want to see what we ask for? Check our developer docs: https://developers.applovin.com/en/ecommerce/events-and-objects/. Anything unexpected gets purged, not stored.
For a deeper technical dive, our CTO’s blog covers these cases in more detail:
How does attribution work?
In apps
We rely on mobile measurement partners (MMPs) like AppsFlyer and Adjust, which are fully integrated with our advertising system. They use IDFA when available or lean on probabilistic matching — tying an ad click to an install via a shared IP in a tight window. IPs shift fast, so no persistent profile forms. They tell us, after a user clicks our advertisement, if our systems deserve credit for the install. If the advertiser agrees to share their data back with us, the MMPs also pass us the activity that happens after an install. Since the installs we measure mostly occur in less than 24 hours, some advertisers have seen incrementality even above 100%, meaning that in those cases we deliver installs which we never received credit for. That’s a proven impact at a massive scale.
In web
We’re just months into web advertising, so we’re still building. Unlike apps, we haven’t fully integrated with third-party attribution firms yet — we use our internal system to report to advertisers. We use first-party pixel cookies and transaction IDs for attribution, not personal info like emails or phone numbers. Apple’s ITP caps their lifespan in Safari, so our web attribution is often swift — 80% of conversions to checkout land within 24 hours. That’s the modern web, and we’re built for it. But ask any client: They rely on their own attribution tools, not ours, to make spend decisions, using last-click or multi-touch models as they see fit. Third-party reports confirm our traffic creates discovery, not cannibalization.
A note for the industry
Today’s performance marketers are analytical powerhouses. They wield multiple tools to measure results and have zero incentive to fund fraud. For many, we’re a major player, collecting over $10 billion run rate annually in verified, valuable spend. If we weren’t delivering, they’d go bankrupt or stop paying — yet our collections are rock-solid. That’s a testament to the sharp minds driving these decisions and the real value we provide.
One final example: the power of learning in action
Now that I’ve covered data and attribution, let’s see Axon in action. Take our first beauty shop client, who is selling makeup. How is it possible we made it work without any knowledge of consumer shopping behavior for makeup? Axon learns fast. Here’s how:
A new ad launches with 500 impressions, scoring a 3% click-through rate — 15 clicks. The model dials back on traits from the 485 uninterested impressions and boosts those tied to the 15 clicks. Some clicks lead to deep site engagement, so Axon finds more like them. The reinforcement loop kicks in: The next 500 impressions hit a higher click-through rate and better engagement. Round by round, it maps our data to new outcomes until sales flow. Sound familiar? TikTok’s algorithm is astonishingly good and works similarly, nailing a new video’s audience through rapid trial and feedback. That’s our personalization edge — it adapts to any advertiser, fast.
Closing thoughts
Advertising, AI, and privacy are complex — worthy of books, not just a blog. But the core is clear: We deliver results for partners, fueling growth, countless jobs, and consumer discovery of games and products they love — all within the rules. Our edge isn’t data hoarding; It’s world-class tech from a small, brilliant team. History shows small groups can change the world: Instagram’s early crew, Signal, Deepseek, and many others. We’re proud to be in that lineage.
This is for our team, our partners, and anyone following our journey. We’re not here to convince skeptics in a few pages — just to share how we work, why it matters, and where we’re headed.
*Grok 3 was used to support the drafting process of this blog. Final content and conclusions are the author’s own.
