3 Easy Steps to Predicting App Revenue Based on SKAdNetwork Data

Starting with iOS 14.5, Apple introduced a new privacy policy, which requires every app to request access to a device’s IDFA. Over 69% of iPhone and iPad owners deny access to their data.

Lack of accurate identification of the device means there is no opportunity to draw a connection between an install and further user activity. As a result, we see classic tools of mobile marketing analytics virtually lose their functionality – an issue we covered in more depth in our previous review.

From now on, to assess the efficiency of ad campaigns, you have to rely on innovative solutions for confidential and accurate attribution.

One of these is LTV forecasting.

How to Predict LTV in iOS 14.5+

Currently, there are two ways to do this: 

  • LTV forecasting with SKAdNetwork Conversion Value functionality for devices
  • LTV forecasting using predictive models based on available data for ad campaigns. 

Let’s look into each of them.

1. LTV Forecasting with SKAdNetwork Conversion Value 

The conversion value is a number between 0 and 63 sent to SKAdNetwork once the app is installed and then shared by Apple with the attributed ad network.

How the SKAdNetwork Works

Interaction with the SKAdNetwork is based on two methods:

registerAppForAdNetworkAttribution

This method is used to register the app install. It notifies Apple that the install should be recorded as a result of an ad campaign and starts a 24-hour conversion value timer.

updateConversionValue

This method is called when the conversion value is updated. These are values that vary between 0 and 63 and are represented in a binary system. An updated CV is always greater than the previous value. However, each time the conversion value is updated, the 24-hour timer restarts.

skadnetwork with CV update
When using the SKAdNetwork conversion value updates, ad networks may receive their install data with a time lag ranging from 48 hours to 63 days.

The CV helps assess how efficient/valuable each install is. Apple suggests every app developer should compile their own conversion table (legend) and use their own values.

Below you can see an example of such a table based on three values: total purchase amount, user action, and predicted LTV.

decode CV
An example of a conversion table (legend) to decode CVs

Sending LTV Predictions via SKAdNetwork

You can map CVs in a table based on any logic. However, it's important to note that values will not be available until (at least) 24 hours after the CV update.

If you map LTV prediction to the conversion value, Apple will aggregate the collected campaign data and send it to ad networks and analytics systems. After that, you will only need to decode the CV into LTV predictions in your chosen currency and assess each campaign performance directly in the interface of your analytics system.

However, this tool is subject to a number of constraints:

1. The same logic is applied to all ad campaigns, with no flexible settings available.

2. Ad networks can’t make use of this data for optimization purposes, at least until they have proposed their own legend system.

3. High indirect costs make it challenging to implement this option in the app on your own: 

    • Back-end calculation of values (for instance, payments total from an install or LTV prediction) and their transfer to the device.
    • Implementation of in-app logic to send data.
    • Own predictive models posing serious development challenges (early LTV forecasting over a long time horizon based on a limited set of user data)

    4. Time lags in displaying data:

    • Apple starts sending CVs only after 128 first installs.
    • Apple sends CVs to ad networks with a time lag ranging from 24 hours to 63 days.
    • The largest networks (Google, Facebook) allow using only aggregated data which results in more time to collect it.

    At the end of the day, we have a clear and reliable solution to assess the efficiency of ad campaigns, albeit with a handful of challenges and constraints.

    2. Ad LTV Forecasting Using Predictive Models

    The other method involves using machine learning models based on available data to predict revenue from users acquired through various ad campaigns.

    Predictions with ATT Consent

    Handling user data for tracking and attribution in iOS 14.5+ is possible only after ATT (ATT – App Tracking Transparency) consent is granted. According to our research, around 30% of users consent to data transfer.

    When consent is granted, standard LTV predictions apply, which are based on user interaction with the app: the number of app visits, payment history, session duration, etc. This approach builds on a set of models, picked for each app individually.

    You can find more information about these models and standard LTV predictions in our article.

    Predictions without ATT Consent

    If the user does not grant their ATT consent, you can only use anonymous activity and purchase data not linked to ad campaigns and transferred Conversion Value. Let us break this algorithm down:

    1. Based on in-app activities and first purchases). Based on in-app activities and first purchases the most suitable machine learning model is selected. After that, the model makes LTV predictions per device. 
    2. The model analyses the number and the value of CV postbacks.
    3. The model reconstructs the distribution of postbacks by install dates, enabling it to analyze the ad campaign by each day.
    4. The model analyzes user attribution to ad campaigns or organic traffic to determine the number of installs across SKAN campaigns. CV is used as an additional signal differentiating traffic quality for each campaign.
    5. Data from precise LTV predictions provided by users who have granted their ATT consent enrich the model.
    6. The model predicts the total app revenue for a given period (e.g. for 30 and 180 days).

    These calculations are performed for each type of revenue separately: in-app purchases, subscriptions, and ad monetization resulting in a complete LTV prediction. To evaluate the efficiency of your ad campaigns, this prediction can be compared to the total or individual ad spend collected by analytics systems from ad networks.

    This is not a user-specific prediction; rather, it is made across a group of users brought in by an ad campaign. This method does not breach Apple’s privacy policy and is the most effective alternative for assessing ad campaigns on iOS 14.5+ given the long user payback period and different types of monetization.

    Prediction Model Toolkit: Key Takeaways

    Currently, prediction models can:

    • Predict an app’s revenue for a long period with high precision, as an early prediction within 12 hours of the install is not required.
    • Distinguish between the revenue generated by organic users and users brought in by ad campaigns.
    • Make an LTV prediction for each type of revenue separately: in-app purchases, subscriptions, and ad monetization.
    • Determine the app install date via the CV receipt date.
    • Allocate the app’s ad revenue to ad campaigns – and consequently to partners.

    As a result, we end up with a simple and convenient tool for analyzing future ad revenue by campaign/partner, which enables the marketing team to streamline ad campaigns.

    Using LTV Predictions for Campaigns with SKAdNetwork Data in myTracker

    myTracker’s predictive analytics system automatically forecasts LTV via SKAN conversions sent by Apple. We have different prediction horizons – 30 and 180 days. All this is available with the optimum (free) plan.

    The process of managing LTV forecasts in myTracker is very simple – you can see your forecasts next to other campaigns in any report in a few clicks (see below). 

    To start working in myTracker you need to register and make sure the app is active and the data is fed to the analytics system (it is necessary to train predictive models).

    Next, follow these three simple steps to build reports containing predictions:

    Step 1: Select metrics in Report Constructor

    Go to Report Constructor and select metrics:

    1. Date
    2. Partner
    3. Campaign
    4. Campaign cost
    5. Installs
    6. SKAN conversions
    7. LTV prediction based on 1m/6m SKAN
    setting up ltv forecast


    Step 2: Group the results

    Go to Report Settings and group the results by campaign:

    set up ltv forecast


    Step 3: Create the report

    Click on the Calculate button – your report is now ready:

    ltv forecast based on skadnetwork data

    Forecasting app revenue enables you to assess ad campaigns by Customer Acquisition Cost (CAC) together with the LTV prediction, and, hence, ROI, even despite iOS 14.5+ restrictions. This helps benchmark different ad sources in terms of their efficiency and optimize advertising costs.

    Please note that not all ad campaigns gather enough data right after being launched. Give your campaign some time to build momentum, and the model will have more data to give you a more accurate prediction. Do not shackle yourself and your business – predict your app’s revenue and assess ad campaigns with myTracker's free predictive analytics.

    Contents