How Yandex Games increased user engagement by 10,5% with AppMetrica’s LTV Predictions

In a quest to increase user engagement, the teams at Yandex Games and AppMetrica developed a predictive analytics tool that forecasts user engagement and gauges its impact on future revenue growth. Learn about the collaboration’s incredible results in this case study.

The challenge

Yandex Games — a mobile app developer and publisher — was looking for ways to improve user engagement and revenue from ad views.

A longtime user of AppMetrica, Yandex Games had previously optimized their ads for an event which was triggered if the user spent more than 10 minutes in the game.

But this way of optimizing ads wasn’t efficient enough. The team had to dedicate a significant amount of time to tracking, observing and fixing the campaigns after they’ve already been launched. This meant ad money was being spent while no action for optimization could be taken.

The solution

While Yandex Games optimized their ads based on the users’ first actions in the game, AppMetrica created an experimental predictive analytics model called LTV & Churn Predictions.

The feature observes how much time a user spends in similar apps and forecasts how that will affect future revenue growth of the app in question. So, to help Yandex Games improve their engagement rates, an experiment was launched.

Comparing optimization methods with an A/B test

Yandex Games became one of the first AppMetrica customers to test the predictive model. To compare the effectiveness of their traditional and new optimization methods, the team ran an A/B test and evaluated which method attracted a more engaged audience.

The A/B test had two options:

1. Optimization based on the «User played for at least 10 minutes» event
2. Event-based optimization based on the predictive model for the top 20% of paying users (top20LTV)

The campaigns were identical in terms of settings and budgets. The first ten days of the campaigns were spent learning and building up the data set, and after that, AppMetrica experts spent a week collecting installation data to use in the predictive model.

About LTV & Churn Predictions

Available now, AppMetrica’s predictive analytics model is a fully developed feature called LTV & Churn Predictions.

The LTV Predictions enable app owners to maximize on their best-performing channels and save money as early in the campaign as possible. The model evaluates each and every new user (anonymously) and sends events to AppMetrica for those in the top 5%, top 20% or top 50%.

Just like how time machines need a destination date, the predictive model requires certain data before it can be launched:

• Installations of similar apps
• User activity the first day after installation

The model is trained using this data and looks 28 days into the future. As a result of the analysis, you will know how much revenue your app will get from user ad views over the next month and optimize your ad campaigns based on those predictions. The cherry on top: all this information will be available to you from Day 1 of your campaign.

The results

Based on the results of the A/B test, the predictive model attracted more engaged users without changing their cost. App usage time increased by an average of 10,5% — meaning that the predictive model is attracting a more engaged and loyal audience.

Moreover, the results were evaluated in terms of how much money the app earned during the first few days of use. Ad views from users who were attracted to the model generated more revenue in the first week after installation than views from users who were attracted to the previous approach over the same period. This trend continued in the following two weeks.

The table below shows the changes in the metrics of the new strategy based on the LTV predictions relative to the default strategy over four periods.The metrics stabilize as soon as the seventh day.

The predictive LTV model can be applied to any app that can be monetized, including games, ecommerce, and apps with paid subscriptions.