Customer Lifetime Value (CLV) and Its Application in Retail Industry

What is CLV?

Customer Lifetime Value (CLV) is an estimate of the net profit a customer will generate over the course of their entire relationship with the company. This helps businesses understand how valuable a customer is over time and guides their marketing and sales strategies. Using data analysis and machine learning methods, the aim of the product is to enable businesses to predict long- or short-term customer values.

How Does CLV Work in the Retail Industry?

In the retail industry, CLV is calculated by considering the total revenue a customer generates over the course of their relationship with the company. The following steps are generally followed in this calculation:

1. Data collecting

Transaction History: Data is collected about customer transactions, including information such as purchase frequency, average order value, and customer demographics. This data is obtained through integration from various sources such as CRM systems, transaction databases and web analytics.

Customer Metadata: Demographic information of the customer such as age and city is collected. These data allow more accurate and detailed analysis.

2.Segmentation

Customers are divided into different segments based on their purchasing behavior and demographic characteristics. This helps tailor marketing strategies more effectively.

3. Calculations

Average Purchase Value: Total revenue / Number of purchases.

Purchase Frequency: Number of purchases / Number of customers.

Customer Lifecycle: The average number of years a customer continues to make purchases.

CLV Calculation: (Average Purchase Value) * (Purchase Frequency) * (Customer Lifetime).

4.Forecast Models

Predict future purchasing behavior using historical data. Machine learning models improve accuracy by taking into account various factors and their interactions.

5.Analysis and Action

Identify high-value customers and focus retention and acquisition efforts on customers with similar profiles. Adjust marketing strategies based on insights from LTV analysis.

Technical Implementation of the CLV Model

Data Preparation:

Data integration is done from various sources such as CRM systems, transaction databases and web analytics. The transaction data provided should include your customers' past transaction information. While providing this data, transaction id, customer id, transaction date and product id information are needed. Additionally, if you want it to be used in modeling, you can provide customer demographic information such as age and city, either on a monthly basis or as a row for each customer.

After providing this data, you need to make some settings for the modeling phase. For example, you may not want to use all of the data you provide and can make settings to use training data from before a certain date. You can choose whether you want feature engineering to be done by looking at how far back your customer's current month is, or you can determine how many months of lifetime value you want to calculate for your customers.

You can use RFM analysis to segment customers who make one or fewer transactions. For example, customers who perform 5 transactions or less can be divided into segments with this analysis, and the lifetime value of the customers assigned to these segments can be examined according to the determined segments.

Models Used:

There are three types of model approaches when estimating CLV. First, we model the CLV value in the coming months with a regression model, with the features we create by looking at your customers' current month and past months according to the value you choose, and the values ​​you choose depending on how many months of lifetime value you want to calculate for your customers.

In the second approach, we again provide a regression model with the statistical approaches provided by the lifetimes library. Here, an RFM analysis is also performed before the prediction and provided as input to the model. In the third approach, we make a multiclass classification prediction.

Model Training and Evaluation:

First, we need to cluster your customers. For example, by looking at the average CLV values ​​of your customers in the first month, we divide them into classes as many as the cluster you choose. If you choose 5, it separates your customers into clusters from clv group 1 to 5, according to their values ​​for that month. We offer quantiles methods for the clustering method. After clustering is done on your existing data, the classification model predicts which clv group your customers will be in in the coming months. After making these settings, you can make your B2metric Automl settings.

In the tab that opens, you can select the models you want to train, provide hyperparameters to increase model success metrics, determine the impute strategy if there are missing observations in your data, and determine validation strategies. After selecting your settings, you can start training the models by clicking save.

Prediction and Interpretation:

Current Shares of Customers and Value

By dividing your customers into classes from the available data, you can see how many of your customers are in which CLV group and the current lifetime value of that group. For example, although there are many customers in CLV group 1, their lifetime values ​​are very low; However, when we look at CLV group 5, you can make comments such as, although there are few customers, their current lifetime values ​​are quite high. You can also look at the estimated distribution of your customers in future groups.

Average Monthly Value per CLV Group

You can view statistical values ​​such as maximum, minimum and median of CLV values ​​in the groups into which your customers are classified.

Evaluation of CLV Groups Between Current and Predicted

With the confusion matrix, you can see the customers that we predicted to be in CLV-group 1 but are actually in CLV group 1, and the customers that were predicted to be in CLV group 1 but are in CLV group 2.

Distribution of Cluster Transition

You can view the number of times your customers switch between groups. The number of people switching from CLV group 5 to 5 is quite high. You can make comments such as there are many customers who have the potential to move from CLV group 1 to 2 and become better customers.

Average Future Monthly Value in the Forecast Window

You can see CLV increases depending on how many months of lifetime value you want to calculate for your customers.

Shares of Current CLV Groups

You can see the CLV group distributions of customers in your current data by month.

Predicted Share of Customers in Each CLV Groups

You can see the expected CLV group distributions for the coming months.

Future CLV Cluster Classification Modeling

You can view the classification model's metrics and see its distribution in future CLV cluster predictions. At the same time, you can view the importance ranking of the features of the trained model and interpret the results of the model by using the confusion matrix and validation data.

Performance Summary

You can evaluate the predictions of three trained model approaches. After the regression models make predictions, they are divided into CLV groups according to the predictions and you can compare the scores according to metrics such as f1 score, precision, recall through CLV groups.