As part of the upcoming WN St. Petersburg’21, AppsFlyer will announce the launch of an open beta version of its predictive analytics solution for iOS. It is called PredictSK.
PredictSKWhat is the essence of the tool?
PredictSK is able to assess the potential of users based on their actions in the first 24 hours after installing the application. Then, based on the received data, he predicts what benefits the advertiser will receive from them by day 30.
PredictSK — response to the appearance of SKAdNetwork
PredictSK is a predictive analytics tool. We can say that this is a response to recent restrictions resulting from Apple‘s special attention to user privacy and, as a result, the deployment of SKAdNetwork in iOS 14.
Recall that among its key limitations:
- the ability to measure user activity only during the first 24 hours after opening the application;
- only one value is sent as part of the postback;
- there is a limit on the number of different events contained in the postback.
Now, when using a classic set of tools, the developer cannot evaluate, including the final LTV, since he is forced to focus exclusively on the user’s early actions.
These limitations can only be circumvented by resorting to machine learning capabilities such as predictive analytics. PredictSK is built on the latter.
How does PredictSK work?
PredictSK uses a 24-hour measurement window provided by SKAdNetwork to application developers. However, as part of the postback, it does not receive a single conversion value, by no means. It allows, thanks to machine learning algorithms, to generate a lot more — information about all measurable interactions during these 24 hours and a forecast for them.
The data obtained is distributed across three KPIs: retention, engagement and monetization, which reflect the LTV logic of a particular application (PredictSK integration requires a short training period during which this logic is analyzed). All three KPIs are rated on a scale from 1 to 9. As a result, a general anonymous assessment of the benefits of each user for the application is given.
By the way, the learning process continues throughout the entire time. New data helps to constantly check and improve the model that evaluates the entire company. Marketers receive it in the form of simple practical information that allows them to optimize the campaign and distribute marketing budgets accordingly.
***
In general, PredictSK’s task is to — taking into account the limitations of SKAdNetwork — let marketers understand how this or that campaign will be good or bad.
A little more about the service will be available tomorrow, July 9, during the AppsFlyer session at the WN Conference.