When Apple launched AppTrackingTransparency (ATT) in 2021, access to deterministic identifiers fell sharply as roughly 80% of users opted out of tracking. User-level feedback loops became sparse and biased, and iOS performance marketing shifted into a different measurement environment under SKAdNetwork (SKAN). Apple’s AdAttributionKit (AAK) later delayed postbacks, compressed conversion values and set privacy thresholds that made signal availability dependent on campaign structure and scale.
What followed was not a smooth transition to aggregated attribution but a market split in how companies responded to the loss of IDFA resolution and the broader move toward privacy-preserving measurement.
The probabilistic measurement phase and structural divergence
In the immediate aftermath of ATT, parts of the ecosystem tried to preserve user-level resolution through probabilistic modeling, using IP addresses and device characteristics to approximate individual matching. For some companies, the possibility of partially recovering user-level continuity reduced the urgency of redesigning systems around aggregated attribution. But that bridge was inherently fragile: Probabilistic attribution remained less accurate and shorter-windowed than deterministic matching. Privacy features such as iCloud Private Relay weakened IP-based matching in web-to-app flows.
Other players, meanwhile, moved earlier toward SKAN-centric optimization, investing sooner in conversion-value strategy, campaign design and prediction under partial observability.
That divergence influenced how models were trained, how campaigns were structured and how organizations learned to operate with latency, sparse feedback and noisy outcomes. The advantage of adapting earlier was first-mover exposure to the operational and modeling discipline required to make that data usable, including choosing which events to encode, structuring campaigns to maximize signal density, calibrating models to delayed feedback and building workflows that could optimize without user-level continuity.
What has driven iOS performance over the last five years
After ATT, advertisers and platforms that adapted quickly by rebuilding optimization around SKAN’s delayed, compressed signals were able to recover efficiency and scale. Those that moved more slowly and operated with weaker postback coverage, less effective conversion-value schemas and slower optimization loops remained materially constrained in measurement quality, optimization precision and downstream performance.
Gaming advertisers were the earliest and clearest pressure test, because they depended heavily on dense feedback, fast iteration and downstream event visibility. But the same shift increasingly mattered for non-gaming advertisers, especially those using iOS for outcomes such as registration, subscription, purchase or activation.
In both cases, conversion value design became strategic because model input quality depended on which post-install behaviors were preserved within a constrained signal space. Campaign structure also took on a measurement role by influencing privacy-threshold eligibility and signal density. Teams that adapted earlier – for example, LinkedIn – were better positioned to turn those constraints into usable optimization inputs.
Four plausible future scenarios
The next phase of iOS performance will depend largely on how far Apple chooses to push AdAttributionKit.
- Apple keeps things broadly as they are.
Adoption of AAK should continue to grow at a steady, linear pace. More advertisers, networks and platforms would gradually shift budget and operational focus toward privacy-preserving measurement, but without a major forcing function, the transition would remain uneven and relatively slow.
- Apple adds more significant capabilities to AAK, but does not materially change its position on probabilistic modeling.
This is the most likely scenario. Apple has already been expanding AAKʼs scope and flexibility, which suggests the company is more likely to strengthen its privacy-preserving measurement rails than to rely on a dramatic new enforcement event alone. AAK adoption would accelerate, and competitive differentiation would increasingly come from what platforms build on top of it: better signal interpretation, campaign design, model training and optimization systems.
- Apple makes probabilistic modeling even harder to use in practice.
A stricter technical or enforcement posture would further reduce the viability of alternative approaches and push more of the market toward AAK. Adoption would likely grow more significantly, along with demand for tooling and expertise built on top of privacy-preserving attribution.
- Apple does both: expands AAK significantly and makes alternative methods harder to use. This would be the most consequential outcome. AAK adoption could accelerate sharply, and the advantage would shift even more decisively from access to identifiers toward expertise. In other words, how well platforms encode signals, interpret postbacks, combine attribution with contextual and operational inputs and turn all of that into better bidding and budget decisions.
Across all four scenarios, the question is no longer whether privacy-preserving attribution will matter, but how quickly the market will consolidate around it and where differentiation will sit once it does.
If Apple mainly expands AAK, platforms will compete on what they can build on top of it. If Apple also makes alternative methods harder, that shift will happen faster. Either way, the long-term advantage is likely to come less from recovering user-level continuity and more from operating better within privacy-preserving measurement itself.
