Modeling attribution on iOS: what works, what doesn’t, and how to choose

SKAdNetwork (SKAN) emerged as a significant shift in the app industry, transforming mobile measurement partners (MMPs) from attribution providers to analytics tools by limiting data access to enriched postbacks from ad networks. This change prompted the development of probabilistic attribution models, offering real-time data without delays, but often inflating performance metrics. Major platforms like Meta, TikTok, Google, and Snapchat have introduced their own solutions, such as Meta's Aggregated Event Measurement (AEM) and TikTok's Advanced Dedicated Campaign (ADC), each with varying degrees of accuracy. These platforms leverage extensive user data to enhance app promotion while maintaining privacy standards. To navigate the complexities and discrepancies of probabilistic attribution, advertisers can employ strategies based on company size and budget, using either baseline uplifts for single-channel campaigns or a four-layer data framework for multi-channel advertising. This approach integrates deterministic SKAN data with probabilistic models, allowing advertisers to estimate real campaign performance more accurately by cross-referencing multiple data points and applying adjustment factors.