Dataseat is now first-to-market with our innovative ML-driven optimization capabilities for SKAN campaigns. These new optimization techniques are already delivering impressive results on SKAN-only campaigns for a major mobile advertiser, including 25% lower CPIs and 64% increase of IPM (install rate metric: installs per mille).

Dataseat – again – proves to lead the game for ID-less performance advertising on iOS. Read on and find out how we are enhancing manual campaign optimization with ML-driven automation for clients’ SKAN-only campaigns.

SKAN: optimization challenges for performance marketers

For anyone even slightly familiar with SKAdNetwork and how limited its signals are, UA campaigns based solely on SKAN data sounds almost like an oxymoron. 

Data loss is enormous and despite Apple introducing new features with every version of SKAN (with SKAN 4.0 being the most advanced and usable so far), the new privacy paradigm leaves very little room for UA managers to drive campaign performance. 

From missing data due to unmet privacy thresholds to navigating complexities in SKAN conversion value management and dealing with very limited data windows, challenges are real. With the anticipation of the fingerprinting ban, app marketers are limited in options for informed user acquisition on iOS, and are turning to contextual targeting and SKAN measurement for their campaigns. It is still a challenge for many how to reach campaign KPIs and get positive marketing ROI in this new reality.

Dataseat aims to marry contextual and SKAN and allow advertisers to learn and improve performance of their mobile UA campaigns with time. 

While working with SKAN data points and their inherent granularity limitations, it’s essential to acknowledge the initial challenges. However, these constraints should not impede our pursuit of data-driven insights. By applying rigorous data science and statistical methodologies on contextual signals effectively, we can navigate these hurdles and harness their full potential for robust SKAN сampaigns model training.Dr. Matina Thomaidou, VP of Data Science, Dataseat

Dataseat’s data science and development teams are dedicating a great share of their time these days solving this targeting problem for mobile marketers and applying machine learning techniques to enable automatic campaign optimization based exclusively on SKAN data. 

Step 1 is the ability to find contexts where the install rate metric (IPM, installs per mille) is the highest, and targeting those.

ML and SKAN-only user acquisition campaigns: first results

The first phase of the project is now complete, allowing automated campaign optimization based on SKAN data only. The first campaigns for a major mobile advertiser went live on September 14, and the results are exciting.

Looking at the campaigns using the first SKAN-based IPM model, we see significantly improved SKAN IPMs and CPIs – 25% CPI Decrease, 64% IPM Increase (based on Dataseat’s SKAN Exploit Strategy Items performance to compare apples-to-apples).

Drilling down on contextual signals

Given that Dataseat is the contextual mobile DSP, we have historical data on the significance of every contextual signal registered and used for campaigns running through our pipes. 

While signals are always rated per individual advertiser and campaign, we are also able to zoom out and look at signal weight across all campaigns, advertisers, verticals etc. Сertain contextual signals have an outstandingly higher impact on ML model training outcomes than others. This core group of signals is more often key to campaign performance, while others frequently help build incremental value.

Narrowing down to those core contextual signals first allows us to check early on for a specific advertiser whether those frequent core factors are indeed helping to reach campaign targets. This kind of prioritization helps check the strongest hypotheses first, while not stopping us from looking at each case individually and tailoring campaign learning to each specific advertiser case.

With the granularity of data being low due to the nature of SKAdNetwork, we have to change our approach from ‘learning loads of things all at once’ to ‘learning more restricted things more sequentially.’ We start with the contextual signals which are historically the most impactful, yet work with each advertiser and their specific needs to customize the learning process accordingly.” Rich Jones, Director of Product, Dataseat.

While data points are still scarce due to the restrictive nature of SKAN, to our knowledge this is the closest the industry can get to using ML techniques for efficiently optimizing SKAdNetwork-informed UA campaigns.

Get in touch and get started with SKAN 4.0

This development couldn’t be timed any better. The impending fingerprinting ban on iOS has been one the biggest elephants in the mobile UA room so far. Dataseat is choosing to help advertisers address the issue by looking for a solution that will allow optimizing UA campaigns in the reality where the only data we have on iOS is the SKAN data.

Curious about where it takes us? Follow Dataseat’s developments as we share them on our Linkedin page.

Let’s talk! To learn more about this experiment and how we work with SKAdNetwork to optimize UA campaigns on iOS, get in touch with Dataseat