Much has been written about Apple’s SKAdnetwork and its component parts, but the Conversion Value is probably the most complex piece that the industry is trying to get to grips with. Ensuring your app is using the optimal Conversion Value definition is key as it’s crucial for two main use cases:
- A value for Ad networks to optimize campaign performance towards, that most closely represents the long term value of a user
- The optimal event to be leveraged for Probabilistic Attribution to be able to rebuild campaign ROAS reporting
However, there are constraints on how the Conversion Value can be crafted:
- how Apple allows it to be defined
- ad network requirements
- MMP support to deliver the Conversion Value model into the app
To optimally define Conversion Value we need to ensure that:
1) the Conversion Value maps early events within the app to LTV (eg D0 KPI to D365 LTV) and
2) we achieve best “clustering” of Conversion Values for Probabilistic Attribution
The Optimal Conversion Value maps D0 KPI’s to long-term LTV (eg: d180 or d365)
It’s impossible to provide a general recommendation because the optimal Conversion Value is app specific, advertisers should use the early events that best correlate to monetization and LTV curves vary across apps.
AlgoLift’s Conversion Schema Algorithm:
As part of our iOS14 campaign measurement solution AlgoLift have built an algorithm that determines the optimal ConversionValue schema for our customers' apps. This model requires 3 data inputs:
Based on the LTV target of your app (eg D180 / D365) and monetization target (IAP/adrev/subscriptions or any combination thereof) the model outputs the optimal schema following MMP and ad network constraints. The model automatically determines which Conversion Value model type is optimal - revenue, event based etc.
We then provide to our clients a static definition for Conversion Value to input into their MMP to be deployed into the app - an example is below.
Manually defining the optimal Conversion Value is almost impossible to do - it’s simple to devise a heuristic but an algorithm is better suited to understanding what D0 events map to long-term LTV (eg d180/365)