Let’s discuss the concept of statistical significance, and make LTV predictions using statistical methods. We’ll discuss when it’s not enough and when it’s worth using Data Science:
- Make a similar LTV prediction using DS methods
- Analyze the nuances of predicting churn
- Learn how to make decisions based on predictions with the help of churn models.
- Using examples, we will explain how customer segmentation together with LTV and churn prediction can help to optimize business.
Customer relationships are important to every company and play a key role in business growth. One of the most important metrics in this area is customer lifetime value (LTV) – a prediction of the net revenue associated with all future customer relationships. The longer customers continue to use a company’s products, increasing profits, the higher their LTV.
There are many marketing articles about how important LTV and customer segmentation are. But as a Data Scientist, I’m more interested in formulas and want to understand how the model actually works. How do you predict LTV using only 3 attributes? In this post I will show some models that are used for marketing customer segmentation and explain the math behind them. There will be a lot of formulas, but don’t worry: everything is already prepared in the Python libraries. The goal of this blog is to show how the math does all the work.
Estimating the accuracy of the model
The traditional approach suggests dividing the data into two groups–part for training, part for testing. In the articles, the authors show that their approach works well. I tried these models on real data, too, and got similar results.
How to apply
As I said in the beginning, all the models are already implemented. For example, the Python “lifetimes” library contains all the functions and metrics needed to determine LTV. The detailed written documentation contains many examples and explanations. There are also sample sql queries to get the data in the required format. So you can get started in just a few minutes.