At the growth stage of every app, company is focused on acquiring as many customers as possible. But after a point, it would not be possible to acquire more customers. That is where a marketer will think of retention.
This is just not for a startup; even the largest of the companies struggle with churn.
As stats suggest, acquiring a customer costs 5 times more than retaining the existing ones. It is very clear now that the main focus should be not only on acquiring new customers but to retain the existing customers as well.
“Prevention is better than Cure”
Keeping this in mind and looking at the user behavior, Raman helps identify users who will not be using your app. This is where you can go ahead and target the users to make them use the app and stop them from uninstalling your app.
Churn Management is an end to end solution Powered by Raman, to predict and arrest churn.
To start making the most of Churn Management, team Raman will help you setup the prediction period. Prediction period is the period for which Raman will predict the users who are likely to churn.
Raman can predict the users who are not likely to use your app for 7/14/28 days.
Once the prediction period has been set, there will be predictions generated for the users who are likely to churn or not use your app in next 7/14/28 days.
These predictions will be divided into 3 buckets viz. “Most Likely to churn”, “Moderately Likely to Churn” and “Least Likely to Churn”. "Overall likely to churn" is aggregation of the 3 micro segments.
Each of these buckets is actually a segment of users who are likely to churn. On each of these segments, there will be a count of users that are there in the segment. There will also be a prediction of how likely are users to churn out if there are no efforts done to stop them from churning.
Once there are predictions of the users who are likely to churn, the next step would be to target these users via some campaign/s.
For that, the user will be able to create a campaign to target a set of users from Churn Management dashboard. User will need to select what segment he wants to do campaign and what type of campaign. It is possible to create Email, APN and SMS campaigns from the dashboard.
Once the predictions are generated and campaigns are done to target users in the prediction period, the next predictions will be generated for next prediction period and the actual performance of the last prediction period will be displayed in the table at bottom for each segment under "past performance".
Each row will be for a prediction period and will show performance metrics like Users predicted, Users Churned, Users retained, the actual churn (predicted plus who were not predicted) and the retention uplift (the percentage lift from users retained through targeted campaigns on predicted segments).
You can further drill down on the performance by clicking on the view more option on each period. Giving you the impact of targeted campaigns (retention people who were target against who were not) and also the monetary impact describing amount of revenue the churned or retained uses gave you in the last 60 days.
You could also analyze you current segments on the dashboard, click on the "Current segment analysis" on the micro segment in question and get a view of the monetary value of these users and also the drill down on particular attributes.
Clicking on the behavior dashboard on a micro segment directs you to the dashboard to analyze your segments even better with user and event level analysis
Raman’s App churn predictions is an advanced neural network based Machine Learning model that takes as input a variety of data from CEE to generate intelligent predictions of app users that are likely to churn from your active user base and thus helping you prevent new churn.
It is important to note that churn here does not mean “Uninstall” of the App but the non usage/interaction of the user with the App as it is more often the case that even if the user does not uninstall the app, he or she does not use it even though the application essentially remains with the users.
Non-Usage/Dormancy is important for tracking and predicting as often this has been seen as a precursor to an actual uninstall happening after some time and it is important that we try and get the customer back with intelligent arrest strategy as soon as we get a hint of dormancy.
Inputs to the model
The App churn prediction model takes in input of User-level data to identify the behavior of each user's usage pattern on the app.
The Different layers to the model and the data used are as follows
User behavioural data profile: This includes data such as frequency of App launch, average session time, all different activities a user performs, etc. The key idea being to identify the user's interaction with the app.
User context data profile: This includes user device, OS etc. The key idea being to create a base user profile.
User engagement data profile: This includes each user's engagement across different channels like Email, SMS, App-Push, and Web-push. This helps the model to generate a holistic view of the user.
The weightage to each layer is optimised by the workings of the model itself.
After this data is given as an input to the Machine Learning model, the model generates a score against each user on how likely the user is going to churn. Inference being users with a high score have a high likelihood of churning, and users with a low score are not likely to churn.
Using this score we divide the users with a high score into 3 segments such that they can be then acted upon to target and arrest the churn:
Users Most likely to churn
Users Fairly likely to churn
Users Least likely to churn
Updated 2 months ago