Shenzhen HiSpark Ad Media Co., Ltd.One-line conclusion: AppLovin has never sold ad slots. It sells prediction. The bet is that it can judge which user will spend money more accurately than the advertiser can.
That judgment explains its most counter-intuitive moves: building its own game studios to validate the buying model, then opening that model to every advertiser, then selling the games outright once the model matured.
This piece breaks AppLovin down across nine dimensions: company basics, the two-wheel model, the algorithm, revenue structure, product matrix, the strategic pivot, competition and controversy, what it means for teams running performance products, and a short summary.
1. Company basics: from a game studio to a mobile ad platform
| Stage | Timing | What happened |
|---|---|---|
| Start | 2012 | Entered through games, solving its own user-acquisition problem first |
| Platformisation | 2015 to 2020 | Productised its internal buying capability into user acquisition and monetization services for others |
| Listing | 2021 | Listed on Nasdaq and entered the public market |
| Expansion and focus | 2021 to 2025 | Bought attribution and connected-TV capabilities, then sold the games business and went all-in on advertising |
Four milestones in AppLovin's path
The takeaway: AppLovin's path was use it yourself first, sell it to others second. It did not get into advertising because it liked the industry. It got in because buying users for its own games was expensive, so it turned that buying into a product.
2. The two-wheel model: games validate, the platform earns
| Business | Role | Value to the other leg |
|---|---|---|
| Game studios | Build and publish in-house | Supplies real buying scenarios and user data to validate the model |
| Ad platform | Serves outside clients | Sells the validated model to every advertiser |
Two wheels: games validate, the platform earns
Core judgment: the in-house games are not a business, they are a test range.
The hard part of any buying model is not compute speed, it is where the labels come from — who actually paid, and who merely downloaded. Owning the games gave AppLovin the cleanest closed-loop data, so the model could iterate on real scenarios. Once the model got good enough, it could monetise independently of its own products.
That is also why selling the games later was defensible: a test range loses its value once the model matures, and keeping it drags on margin.
3. The algorithm: turning media buying into a prediction problem
Traditional buying thinks in auctions: whoever bids highest wins the impression. AppLovin thinks in predictions: estimate how much revenue this user will generate, then work backwards to what you are willing to pay.
| Dimension | Traditional buying | Predictive buying |
|---|---|---|
| Core question | What is this slot worth? | What is this user worth? |
| Basis for decisions | Historical averages, audience labels | Predicted value of an individual user |
| Data required | Impressions and clicks | The full chain from impression to payment |
| Team shape | Media ops heavy | Algorithms and data heavy |
From auctions to predictions
The implication is counter-intuitive: the stronger the algorithm team, the fewer media-ops roles you need. That is why the headcount profile of a company like this looks nothing like a traditional ad agency.
4. Revenue structure: software overtakes games
| Stage | Rough shape | What drove it |
|---|---|---|
| Games-led | Games were the bigger share | In-house titles generated the top line |
| Platform catching up | The ad platform grew faster | Advertisers validated the model's performance |
| Platform-led | Software became the main source of revenue and profit | A higher-margin business took a bigger share |
| Focused | Games sold | Resources concentrated on the ad platform |
Four stages of the revenue mix
The margin gap between software and games is the key to valuing this company. Games are a content business whose revenue swings with product cycles. An ad platform is an efficiency business, and once it works it scales.
5. Product matrix: a four-part loop
| Product line | Step | What it does |
|---|---|---|
| User acquisition | Find users | Puts ads in front of the people most likely to pay |
| Monetization aggregation | Sell inventory | Gets developers a higher price for their ad space |
| Attribution | Measure | Works out which channel actually created value |
| Connected TV | Extend reach | Carries mobile capabilities onto the TV screen |
Together the four form a loop: find users, sell inventory, measure results, extend reach. A loop means data does not leak out, and it means an advertiser who integrates rarely leaves.
6. The pivot: selling the games to focus on advertising
Selling the in-house games business is the single most instructive decision this company made.
There were three layers to it. The first is margin structure: software margins are far higher than games. The second is client trust: when a platform is both an advertiser and an ad vendor, other game companies worry that their data is being used to improve a competitor's products. The third is the valuation logic: public markets have long paid a higher multiple for pure platforms than for content businesses.
One general lesson for product teams: when a business turns from validating a hypothesis into maintaining a stock, its strategic value has already moved on.
7. Competition and controversy
| Rival type | How they compete | Pressure on AppLovin |
|---|---|---|
| Giant ad platforms | Take budget with ecosystem and traffic scale | Large-client budgets get split |
| Independent peers | Compete for the same developers | Bidding pushes traffic costs up |
| Developers building their own | Top studios build in-house buying systems | The best clients leave |
Controversy is part of this company too. Short sellers have questioned its data and reporting metrics in recent years, and the company has denied the claims. For an observer this is a reminder that the performance of predictive buying is very hard to verify from outside, so advertisers can only judge by their own conversion data. It also means the cost of trust in this category runs higher than the cost of technology.
8. What it means for us: five questions for a performance product
| Question | How to ask it | AppLovin's answer |
|---|---|---|
| Data source | Do you have clean, real labels? | In-house products supply closed-loop data |
| Nature of the business | Are you selling slots or predictions? | Predictions. The slot is worth little |
| Portfolio choice | Keep the test range? | Sell it once the model matures |
| Client trust | Will you compete with clients? | Sell the games and remove the conflict |
| Cost structure | Can traffic cost fall? | Amortise it through a higher conversion rate |
The last row is the core: in a buying business, the only sustainable way to hold costs down is to convert better than your peers.
9. Summary
- Identity: AppLovin is a predictive buying platform. Its real product is judging the future value of a user.
- Structure: in-house games were a test range, and could be sold once validation was done.
- Path: use it yourself, then sell it to others — the most reliable cold start for a tool-shaped business.
- Risk: performance cannot be verified from outside, so the cost of trust exceeds the cost of technology.
A five-question checklist plus a breakdown of the use-it-then-sell-it cold start.
What is inside:
- The five questions to ask about a performance product
- A breakdown of the use-it-then-sell-it cold-start path
- A self-check list for the monetization architecture of an overseas product
