How Auto.ru doubled its revenue with LTV Predictions optimization

Intro

Auto.ru, a major online car seller, was looking for a way to optimize its mobile ad campaigns by attracting more high-value users without increasing its marketing budget.

Learn how they used LTV Predictions in AppMetrica to find a target audience with the best lifetime value and optimize their campaigns to double its revenue without changing the budget.

The Challenge

In a classic tale of ecom app optimization, Auto.ru wanted to improve its user acquisition strategy to attract more high-value users and grow its revenue without increasing its marketing budget.

As the Auto.ru team tested various optimization approaches, Machine Learning experts at AppMetrica’s partner team called Crypta, developed a predictive analytics feature LTV and Churn Predictions. The feature allows app managers to identify users with the highest lifetime values and where they came from. Then app owners can optimize their UA campaigns to attract similar users.

Auto.ru team members were intrigued. LTV and Churn predictions could potentially allow them to optimize their campaigns without hurting their marketing budget. So, they decided to give it a try.

The Solution

The team has been using AppMetrica to track its ad campaigns. With the platform’s new LTV and Churn Predictions feature, they saw an opportunity to not only see predictive analytics in action, but also discover an effective optimization strategy relatively easily.

AppMetrica’s LTV and Churn Predictions feature helps evaluate user data and make predictions about the best channels and ad campaigns to attract a high-value audience. In the case of Auto.ru, AppMetrica helped understand how interested each user would be in the app in 28 days after installing it and how much money they would bring.

To choose an optimal optimization strategy, Auto.ru and AppMetrica decided to run an experiment to compare optimization based on topLTV and the company’s traditional methods. The teams launched 3 campaigns in Yandex Search and Yandex Advertising Network via AppMetrica. Each campaign had the same settings and budget, except the target goal. The test compared optimization based on the behavior of 3 user groups:

• Top 5 LTV users
• Top 20 LTV users
• Control group, where campaigns were optimized based on the users’ actual actions in the app

The test and measurements took about 2 months to ensure high quality traffic and accuracy of the data.

The Results

The results of the campaigns didn’t leave any room for guessing. Spoiler alert: optimization for users in the top5 LTV group beat every other method.

As the teams compared all three campaigns against each other, they saw:

• Campaign revenue optimized around Auto.ru’s top 5 LTV users was 80% higher than the revenue from users in the control segment.
• Average revenue per user in the top5 LTV group multiplied 9 times compared to the control group
• With optimization based on topLTV, the fraction of revenue-generating users multiplied 10 times.

The company’s overall revenue doubled. At the same time, the number of installations decreased 4.5 times, meaning the traffic that came through the top5LTV campaign was more high-quality and paying than before, and Auto.ru didn’t need as many users to download their app to get the same amount of revenue (or more).

Moreover, the team concluded that quantity clearly does not guarantee quality. By attracting a large number of users (target and non-target), Auto.ru would have to spend money on users who may not bring them any revenue. But by attracting users with higher LTV scores, they get a guarantee that for the money they spend we will generate a profit and be worth the investment.

Now that’s what we call optimization!