Effect of push notifications for wavers

Comparison of wavers addressed via push notifications and users without re-engagement actions.


In our most recent study, we have been working with the team from Zoobe who have been using goedle.io’s Machine Learning API to re-engage their potential churners. Zoobe has developed a quite popular entertainment app delivering character voice messaging with over 7 million installs and more than 200 million created messages. It enables 3D animated characters to actually “perform” a voice message from app users and then creates easy-to-share video clips.


Exemplary churn report

Exemplary churn report for Zoobe with altered statistics. The full report contains more features.


To obtain a better understanding of their user base, Zoobe’s product team first of all utilized our churn reports. These insights helped them to save plenty of time building complex cohorts and descriptive graphs in their analytics provider to actually figure out how their loyal users look like compared to their churners.

The next step was to derive actionable items from these insights to improve the engagement of these segments:

To increase the retention on day-15, Zoobe consumed goedle.io’s wavers-API to determine users that should be approached via push notifications to avoid their churn. goedle.io defines wavers as users that are neither predicted to be very loyal users, nor users that have already churned with certainty. More specifically, goedle.io’s machine learning stack predicts these users to potentially have a promising engagement in the near future.

Over a period of two weeks, Zoobe sent push notifications to wavers on their 15th day after installing the app. One group was targeted at a fixed time, while a group of the same size was targeted at a time determined by goedle.io’s moment-API. This API-endpoint computes the most suitable moment to approach users in a data-driven fashion.

Comparing users that did not receive a message to those users who were addressed, the most significant improvement on the day-15 retention was achieved for wavers that were targeted based on goedle.io’s scheduling – increasing the retention by almost 64%. The following two days thereafter also featured an improvement of retention compared to those users who did not receive a push notification.

While sending push notifications at a fixed moment also improved the day-15 retention, the improvement was smaller than the gains obtained by the data-driven scheduling. This clearly shows that approaching wavers at an algorithmically determined moment is a promising starting point for re-engagement campaigns.

In the future, Zoobe and goedle.io will continue to evaluate other settings where methods from Big Data and Machine Learning promise to simplify and improve mobile marketing tasks. This includes an evaluation of the same setting, however, after observing the users for a shorter period of time after installation of the app. In the extreme case, one is interested in obtaining an indication about churn behavior after the first 24 hours of usage (day-1 retention).

Another interesting setting replaces simple textual push notifications with more tempting incentives such as vouchers or promotion codes. Here, the goal is to find the optimal set of users that should receive a free or price-reduced item. While in this setting the group of wavers still depicts an interesting subset, patrons might be worth targeting as well, to further promote in-app-purchases.

If you are interested in the full technical setup of this use case with data-driven scheduling, you can find a brief description here.