Basic Customer LTV prediction with Python | DataVault
Basic Linear Regression to predict LTV for Ad Revenue Apps
In this use case, we will learn how to predict customer lifetime value (LTV) in python using using decoded data from Tenjin’s DataVault.
What will I be able to predict?
This model is an extension, or an advanced version, of our previous use case on LTV prediction in Google Spreadsheets. In this model, we use Day 0, Day 1, Day 2, and Day 3 LTV data from your current campaign to predict Day 7, Day 14, Day 30 and Day 90 LTV in python. The predictions can be made at a campaign, site ID and app level. For this model, you would need access to more granular data from Tenjin’s DataVault, which is a paid product.
How does the model work?
The model uses decoded data in Tenjin’s Datavault from different Mediation Partners or Monetisation channels such as Applovin and Topon for a simple “on-the-fly” Linear Regression Framework. After performing the regression, the data is pushed back into DataVault. This process can be visualised in the graphic below.
How long does it take to get the predictions?
What does the dashboard look like?
Below, you see an anonymised version of the dashboard. Feel free to navigate through the filters.
What are the benefits of using this model?
- This is a simple model that is easily explainable
- It’s also consistent with the 4-Step LTV prediction in Google Spreadsheets use-case
What are the limitations of using this model?
- Right now this model is only valid for Tenjin’s DataVault customers. However, if you are not a Tenjin customer, and would like to use this model for your LTV predictions, then reach out to us.
- The model is computationally expensive and requires more execution time than our advanced models
- This model is also prone to outlier observations
- The predictions for this model require four data points to be function (Day 0, Day 1, Day 2 and Day 3 LTV)
How can I get started with this model?
If you’re interested in running the python script then reach out to us and we can either send you the script or run it for you.