Predictive lifetime value

Adjust’s predictive lifetime value (pLTV) solution gives you insight into the lifetime value of your users while protecting their data privacy. Combining Adjust's proprietary machine learning model with SKADNetwork deterministic attribution, you get the best of Apple’s user protection and Adjust’s powerful data analysis right inside Datascape.

With the release of iOS 14, Apple introduced SKAdNetwork, a privacy-protecting tool for attributing installs. SKAdNetwork returns limited information about installs to advertisers to protect user privacy, which restricts insight into user behavior. SKAdNetwork postbacks contain a conversion value for each user, which marketers can map to a range of actions in their app. pLTV enables you to build a predictive model of user behavior and map predicted outcomes to conversion values to receive an accurate view of post-install user behavior.

How does it work?

To enable pLTV, we collect anonymized session, event, and revenue data from your app’s users. We pass this information to a machine learning (ML) model to compare a user’s day 0 behavior to their future behavior. Once the ML model has processed enough data, it can accurately predict a user’s behavior on a given day in the future.

Note: The more data we collect, the more accurate the model’s results will be. We need to record at least a few more days of data than the period you want to predict for. For example, to predict 14 days of lifetime value, we need to collect approximately 20 days of data.

The workflow for predicting a user’s day 14 lifetime value (LTV) looks like this:

1. Adjust receives data from the SDK when a user performs an action.
2. Adjust combines this data with other user information, such as device data and previous purchases.
3. Adjust's machine learning model processes this data to predict the user's day 14 LTV.
4. Adjust converts this predicted LTV into a conversion value and sends it to Apple.
5. Apple returns an attributed postback after the SKAdNetwork timer expires. This postback contains the assigned conversion value.
6. Adjust unpacks the conversion value to predicted LTV and combines it with other postbacks.
7. You see the aggregated predicted day 14 revenue for each campaign in Datascape. You can then compare these campaigns to make decisions about future investments.

Generation buckets

One of the limitations of SKAdNetwork conversion values is that their value can only increase. Predicted values may change based on different decisions the user makes while using your app, meaning predicted values can either increase or decrease. To work with this, we map outcomes to conversion value generation buckets. A generation bucket contains a range of conversion values that outcomes can map to. If pLTV predicts a higher conversion value and then predicts a lower one, it can use one of the higher conversion values contained in the generation bucket of the lower value result.

Generation buckets map minimum and maximum revenue values to individual CVs. pLTV uses these buckets to assign in-app revenue events to variable conversion values that can go up or down. In the following example configuration, we map revenue to eight buckets across two generations. pLTV uses the buckets in the first generation initially, and uses the values in the second generation on receipt of a second update.

Here is an example of how pLTV uses the above generation buckets to update SKAdNetwork:

1. A user installs your app. This updates the conversion value to 0 and sets a predicted revenue of $0.
2. The user performs an action in your app that increases the predicted revenue value to $1. Adjust sends CV 1 to SKAdNetwork.
3. The user performs more positive actions and their predicted revenue increases to $7. Adjust sends CV 3 to SKAdNetwork.
4. The user performs some negative actions that decreases the predicted revenue value to $5. Since Adjust cannot send a CV lower than 3, it sends the CV from generation 2 that maps to $5, which is CV 6.
5. Apple sends a SKAdNetwork postback with a conversion value of 6. Adjust displays a pLTV reading of $5 for the user in Datascape.

If you’re ready to get started with pLTV, contact your technical account manager.