Acquisition Battle #1: SKAN vs Probabilistic Model

Mobile attribution has undergone significant changes in recent years, largely driven by Apple’s App Tracking Transparency (ATT), which has reshaped data collection and analysis practices. Within this new framework, two main approaches have emerged on iOS to address advertisers’ challenges : SKAN Vs Probabilistic Model

On one hand, SKAN, developed by Apple, prioritizes user privacy with a strict yet universal framework for iOS campaigns. On the other, the probabilistic model has gained traction as an alternative, designed to offset some of the limitations introduced by ATT.

Before We Dive In

To better understand the fundamentals of these two approaches, we recommend exploring these detailed articles:

• All About SKAN
• The Probabilistic Model Explained

Now, what are the strengths and limitations of each model? Let the battle begin.

Round 1: Data Accuracy

SKAN

With its privacy thresholds, limited number of Campaign IDs, and delays in receiving postbacks (24 to 48 hours), SKAN makes granular performance analysis challenging. While the anonymity tiers introduced with SKAN 4.0 bring some improvements, these constraints still hinder detailed post-install event tracking, making quick optimizations in user acquisition more difficult.

Probabilistic Model

The probabilistic model stands out for its granularity and speed. With no delays in data transmission and flexible attribution windows via MMP tools, advertisers gain precise visibility into performance at every stage. However, its accuracy depends heavily on the quality of the initial data and algorithms. Poor calibration can lead to biases or over- or under-attribution, distorting performance measurement.

Round 2: Compliance with Regulations

SKAN

SKAN is Apple’s solution to meet the requirements introduced by ATT. By eliminating the use of personal identifiers, SKAN aligns seamlessly with this strict privacy framework, providing advertisers with full compliance with iOS regulations and peace of mind amidst evolving policies within Apple’s ecosystem.

Probabilistic Model

While the probabilistic model is currently tolerated, it is more vulnerable to policy changes. Some methods rely on advertising identifiers that could be affected by stricter rules from Apple. To maintain compliance, advertisers need to stay vigilant and ready to adapt their practices, a process that can take time and impact performance until the necessary adjustments are fully implemented.

Round 3: Data Visibility and Measurement

SKAN

SKAN is a reliable solution for measuring installation volumes, although data can sometimes be slightly underestimated. However, its major drawback lies in in-app visibility, which is heavily constrained by Apple’s thresholds. These thresholds require reaching a specific user volume to generate actionable data, often making granular analysis impossible.

While SKAN works well for campaigns focused on installations, it quickly shows its limitations for more advanced goals, such as subscriptions. As for retargeting, it remains largely irrelevant, making it challenging to extend its use beyond acquisition.

Probabilistic Model

The probabilistic model provides complete visibility, covering both installation volumes and post-install data. Unlike SKAN, it allows granular tracking of in-app actions, such as purchases, delivering crucial insights for campaigns with advanced objectives.

This depth of post-install data analysis is vital for performance optimization, where SKAN struggles to compete. However, precise configuration of attribution windows is critical to ensure reliable data and avoid biases.

Round 4: Implementation Complexity

SKAN

Implementing SKAN can be a real challenge. Understanding technical concepts such as postbacks, coarse and fine-grained values, and attribution window management requires specific expertise. These complexities can make its adoption difficult for less experienced advertisers.

Probabilistic Model

The probabilistic model is often considered simpler to use, especially with MMPs simplifying its management. However, this apparent simplicity comes with its own risks: poor configuration can negatively impact results. Proper management of attribution windows is critical to avoid biases, as incorrect settings can lead to over- or under-attribution.

To ensure consistent and actionable insights, probabilistic data often needs adjustments to align with SKAN’s installation volumes, which are recognized for their reliability.

Conclusion: Two Methods, Two Approaches

There’s no clear winner in this battle. SKAN remains a reliable solution for measuring installation volumes, offering a clear and universal framework that ensures long-term stability amidst Apple’s policy changes. However, it quickly reaches its limits when it comes to tracking in-app performance, where its granularity falls short.

The probabilistic model, on the other hand, stands out with greater precision for in-app performance and advanced goals, delivering detailed and actionable insights for complex strategies. However, in terms of compliance, it’s less stable, with its future tied to market developments. Additionally, improper configuration can distort results, making it essential to handle this model with care and expertise.

 The right approach to attribution may not be a matter of “or,” but of “and.” By combining the strengths of both models, advertisers can build robust strategies: leveraging SKAN’s stability and reliability for installations while utilizing the probabilistic model’s granularity and agility for more ambitious campaigns.