AppsFlyer: How Neutrality Became a Moat

Context

Every mobile advertiser has the same problem. They spend across Meta, Google, TikTok, Apple Search Ads, and a long tail of networks, and at the end of the month each of those networks sends a bill claiming credit for the same installs. Meta says it drove the install. Google says it drove the install. Both are looking at their own click logs and counting last-touch in their own favour. Someone has to settle the score.

That someone is the mobile measurement partner. AppsFlyer is the category leader, and the reason it won is not a clever feature. It is a refusal.

What I noticed

AppsFlyer does not buy or sell media. It has no inventory, no auction, no incentive to make any single network look good. It sits on the advertiser’s side, ingests the SDK events from the app, and adjudicates which touch gets credit using rules the advertiser controls rather than rules the network controls.

That sounds modest. It is the entire business. I own the data behind a DSP and an ad exchange, and I can tell you the dirty truth of this industry: every party that sells media also reports on its own performance, and every party that reports on its own performance inflates it. Not always fraudulently, often just through generous attribution windows and last-touch math that happens to favour the seller. The advertiser knows this and cannot trust any single number. The neutral referee is the only number everyone can agree to argue about.

The product thinking

Neutrality is usually framed as an ethic. Here it is a market structure. AppsFlyer made itself the clearing house that the conflicted parties route through precisely because none of the conflicted parties can be the clearing house. Meta cannot be the source of truth on whether Meta works, for the same reason a striker cannot referee his own offside.

This produces switching costs that have nothing to do with software lock-in. The advertiser’s historical attribution data, their measurement windows, their fraud baselines, their cohort definitions all live in the MMP. More importantly, every network the advertiser works with is already wired to send postbacks to that MMP. To switch MMPs you don’t just migrate a tool, you renegotiate the plumbing with a dozen ad networks at once and lose the historical baseline you benchmark new spend against. The switching cost is the network of integrations, and AppsFlyer’s neutrality is what made every network willing to integrate in the first place.

That is the elegant part. A media seller could never have built this graph, because rival sellers would never wire themselves into a competitor’s measurement layer. Only a party that competes with no one gets universal integration. Neutrality was the price of admission to becoming infrastructure.

The IDFA test

The thesis got stress-tested in 2021 when Apple killed deterministic device-level tracking with ATT. The deterministic install-to-click join that powered classic attribution simply stopped being available for the majority of iOS users. A naive read said the MMP was dead: no device ID, no source of truth.

What actually happened is the more interesting outcome. AppsFlyer pivoted the same neutrality into the aggregate world. It built SKAdNetwork aggregation and conversion-value modelling, a data clean room for privacy-safe joins, and an incrementality and lift-measurement product that answers the harder question of what spend actually caused. The measurement primitive changed from “which click drove this install” to “what is the marginal contribution of this channel,” and the neutral party was still the only one positioned to answer it. If anything ATT strengthened the case, because aggregate signal is noisier and noisier signal makes a trusted, methodologically consistent referee more valuable, not less.

The risk

Here is where I’d push back on my own argument. The single-source-of-truth thesis was strongest when truth was a deterministic row: this device, this click, this install. When signal goes aggregate and modelled, “truth” becomes a methodology rather than a fact, and methodologies are debatable. SKAN gives you a coarse, delayed, privacy-thresholded postback. Incrementality is a model with assumptions. Once the answer is a model, the advertiser can ask why they should trust AppsFlyer’s model over Meta’s own modelled measurement, or over Google’s, or over a media-mix model they build in-house.

The walled gardens know this. Meta’s Aggregated Event Measurement and Google’s modelled conversions are bids to own the modelled layer directly, cutting the referee out by saying “trust our model, it sees more data than the MMP does.” When the unit of truth is aggregate, the party with the most first-party data has a credibility advantage, and the walled gardens have more data than the neutral middleman. Neutrality still beats self-interest as a trust argument, but it is a weaker beat when no one can independently verify the score anyway.

The PM lesson

The most durable moats are sometimes a constraint you impose on yourself. AppsFlyer’s refusal to sell media looked like leaving the biggest revenue pool in adtech on the table. It was actually the thing that made the entire industry route through them. The open question for the next decade is whether neutrality survives the shift from counting facts to modelling them, because a referee is only valuable when there is a real score to keep.