"The earlier you start re-engaging the user with an effective ad displayed at the right moment, the more cost-efficient it will be. Retaining users is cheaper than reacquiring them."
User retention is absolutely crucial.
Unfortunately, after mobile apps acquire users, we observe that most of these users drop off within the seven days following install. It’s a well-known fact that the longer a user stays, the higher the chances of them making purchases or any targeted action, increasing user LTV. User acquisition drives new users while re-engagement maximizes their potential.
For most of our clients, the decision to start re-engaging users is based on the inactivity window, but sometimes that's not enough. In the past, it was a reliable proxy. Nowadays, an increasing number of mobile app marketers are approaching re-engagement differently— to be more precise with timing, specifically when they should start re-engaging their users and begin showing ads to them. The earlier you start re-engaging the user with an effective ad displayed at the right moment, the more cost-efficient it will be. Retaining users is cheaper than reacquiring them.
Churn prediction can be leveraged to combat these issues. First, it complements user acquisition at the right time by reinforcing their interest at the right moment and ensuring their retention within the app. It also provides flexibility in moving away a bit from the inactivity window proxy and finding a more effective point to re-engage the users at the time they should see the ad to prevent them from leaving the app.
Second, churn prediction allows you to foresee potential financial losses if you’re not successful in bringing the users back. Indeed, there's a lot at stake in terms of losses and gains when you undertake re-engagement— it depends on the app and its business model. For instance, with hyper-casual games, the LTV of the user is quite short, and it's acceptable for them to churn early. In contrast, dating apps aren't interested in retaining lapsed users because it means that their users have already found a match; they focus more on activating users and upselling them. But for gaming apps, they invest a lot in acquiring players and want to do everything possible to keep them engaged in the app, making purchases, and increasing their LTV.
Many of our clients, especially the larger ones, are already working on churn prediction. However, it's highly challenging. It requires a lot of commitment and investment. Starting from this year, I’ve seen an increasing number of third-party solutions entering the market to provide churn prediction tools, and Adikteev is one of them.
To make re-engagement efficient, you need the right tools. Even though we make predictions on a user level, the end result is a segment that aligns with the incoming Android Privacy Sandbox. This eliminates the need for our clients to invest their own resources into development.
We thrive on challenges and we’ve conducted head-to-head tests with our churn prediction tool. In the vast majority of cases, we’ve emerged victorious. If not, it provides us with a wealth of knowledge to make our tool even better, to be efficient in what we are doing.
How Churn Prediction Works
As you know, machine learning and artificial intelligence are primarily focused on harnessing as much data as possible to feed the algorithm and provide the outcome. We take ample raw data to analyze user behavior patterns, such as frequency of their activity, last session, session lengths, purchasing behavior— we combine these with monitoring signals emanating from those specific users. Then, we compare this information against known historical data. This enables us to build different user segments with their respective likelihood to churn. Subsequently, you can fully utilize these data-driven insights and take action.
We use this system that allows us to score the probability of users to churn from 0 to 1, where 1 signifies the highest likelihood for users to churn. As a retargeting provider, we, of course, focus more on re-engaging users that are most likely to churn. That's why we strongly recommend concentrating on something that is higher than 0.8 in a churn prediction.
Let’s look at this use case from one of our clients, a game-in-one app.
On these slides, the dark blue columns represent the users that are re-engaged by the current retargeting activities of this specific client. Meanwhile, the green ones represent the segments predicted to churn. This data highlights a group of users not covered by retargeting campaigns and who have the highest likelihood to churn. This revelation presents marketers with ample room to adapt their current retargeting activities. You can either engage with the untouched segment or run dedicated retargeting campaigns for those specific users who will likely churn.
When we introduced our churn prediction tool to the market, we evaluated its validity and reliability. Thus far, we have observed that within the user segments most likely to churn, our tool has demonstrated more than 90% accuracy. This figure was derived from comparing our predictions with actual outcomes. We also believe that prompt action can minimize financial losses that could occur if paying customers were lost.
Adikteev is a pioneer in the use of uplift tests in incrementality. We apply uplift tests to everything we do. We have observed that users in segments with the highest probability to churn are unlikely to return to the app organically. We noted a remarkably strong uplift for these segments, indicating that retaining these users is not only unprofitable but also a poor investment. Our goal is to safeguard the finances of all our clients.
Regardless of how you implement churn prediction, it provides a variety of benefits. Its use extends beyond programmatic retargeting as remarketing spans multiple channels, such as CRM activities and social media. The call to action here is to initiate proactive steps to improve efficiency.
Furthermore, the inactivity window can serve as a benchmark, separating campaigns that target users predicted to churn from those who remain active. Some campaigns aim to re-engage users most likely to churn. This strategy involves extensive analysis.
This knowledge allows you to understand user behavior and align with their patterns, which ultimately influence whether they stay or churn.
In conclusion, here are the key insights regarding churn prediction for mobile apps:
Firstly, taking prompt action at the right time is crucial because we are dealing with future possibilities. This is a common topic we discuss with our clients and peers in the industry. The inactivity window is simply not efficient enough, especially for retargeting. It’s necessary to identify the optimal time for re-engaging users to maximize benefits.
Churn prediction can also yield valuable insights on why your users have churned. As you delve deeper into whether churn prediction is effective, you can apply those insights to User Acquisition (UA) and product development. This knowledge allows you to understand user behavior and align with their patterns, which ultimately influence whether they stay or churn.
Another factor to consider is that user behavior patterns can vary between apps and verticals. Some users may use an app daily, while others may use it less frequently. For instance, some users might typically use an app once every 14 days. Retargeting such users repeatedly within a 14-day inactivity window may not be the most effective strategy.
Churn prediction offers more precise and granular information. As your predictions become more accurate, you can save a substantial amount of money. Financial loss is a critical concern, and eliminating it is key to success.