ローマン・ガーバー
10月 29, 2020
Starting from 2021, advertisers on iOS won’t be able to calculate key cohort metrics, like ROI or LTV, for most of their ad campaigns. This creates a new paradigm where both mobile publishers and vendors will need to innovate to succeed. At Tenjin, we are focusing our efforts on building a new type of methodology for assigning installs to marketing campaigns – Attribution Modeling.
For the past few months, we have worked on an entirely new model for attributing users, in the new reality, for marketing that will allow advertisers to make educated decisions.
Developers need to rethink IDFA based ROI measurement since it will only include 20% of traffic.
Due to upcoming IDFA limitations, deterministic attribution will become obsolete on iOS.
Why won’t iOS advertisers be able to optimize their traffic like they used to?
Mainly for two reasons:
1) Cohort ROI and LTV are possible only thanks to deterministic attribution (based on IDFA), allocating 100% of the credit to a particular click or impression. However, in 2021 deterministic attribution will only be available for users that opt-in. We estimated this number at 20% for SANs back in July. The actual number might be lower since some SANs may not collect IDFA at all – Facebook’s example.
2) SKAdNetwork, by design, does not measure long-term user activity. You can read more about it in this guide.
Simply put, if you want to measure campaign ROI for more than 20% of your user base, you will need to change how you look at your advertising data.
By the end of 2021, we expect the most advanced publishers to make a jump from deterministic to something else.
Attribution Modeling is a new way to look at marketing data
So, what is the difference between attribution modeling and deterministic attribution?
- Deterministic attribution – an acquired user is attributed to one ad source based on the last signal received;
- Attribution modeling – an acquired user is assigned independent probabilities that they came from different marketing sources based on various input signals.
Attribution Modeling sounds different on the surface, but assigning probabilities is not new in our industry. Take, for example, last-click attribution. It’s an industry-wide standard to give full credit (100% probability) for the install to the last click regardless of other signals (impression, other clicks) that took place to produce this install.
In 2021, mobile publishers will need to figure out how to use less determinate signals since last click deterministic attribution requires IDFA. Counterintuitively, the marketer’s role will require MORE data analysis, not less. Mobile publishers will compete in how accurate their prediction models are when understanding where users come from based on indeterminate signals.
Tenjin’s Attribution Modeling will be integrated into existing workflows to enhance the existing products.
Our team is working on a first MVP model that will allow our customers to model attribution using SKAdNetwork data and other sources compliant with Apple’s privacy requirements.
One thing to get out of the way when we talk about Attribution Modeling in Tenjin:
Tenjin’s Attribution Modeling is not a re-skinned version of fingerprinting. It’s an entirely new product.
I feel that I need to explicitly emphasize that the term ‘probabilistic attribution’ is misused a lot now, and many vendors are using it to cover up what is really just fingerprinting attribution. Just to be clear, we are not doing that with our Attribution Modeling. And that’s why we decided to go with the term Attribution Modeling instead of probabilistic attribution.
We strongly believe that just like the deterministic attribution during the IDFA era, Attribution Modeling will become a new standard for advertisers during the SKAdNetwork era.
If you would like to become a first adopter of Attribution Modeling, write us an email at info@tenjin.com.
