Calculating Dynamic Customer Lifetime Value (DCLV) using Data Science

One of the most important topics for marketing analytics is the prediction of customer lifetime value (CLV). Companies can use the CLV prediction to understand an individual client (dynamic CLV) or segment their customers (static CLV) based on their predicted value. It may also be used to improve and prioritize business strategies. Companies across the industry spectrum have been investing millions of dollars in data & analytics to better calculate and predict CLV.

What is Customer Lifetime Value? 

Before we dive into the calculations, let’s discuss the definition of CLV. CLV is the customer’s total profit value to your company throughout the entire time they are a customer. This includes the money they have spent, will spend, and the value of potential referrals to other prospective clients. 

CLV tells you the overall profit expected from a customer and helps you to determine how much you should be investing in them. CLV also provides you with insights about whether the client will be renewing the contract and how much more you could cross-/up-sell to the client. Having an effective CLV calculation helps companies to improve business strategies for attracting new clients, retaining existing clients and maximizing profit margins. 

For example, a subscription-based CRM tool company is interested in understanding the average CLV: the percentage of customers renew the subscription service every year, the percentage of customers that upgrade their product tier to broader CRM services, and the percentage of customers that recommend the tools to others.  

---

Calculating Customer Lifetime Value: Customer Lifetime Value Formula 

Now that we have introduced the definition of CLV, let’s talk about how to calculate it. 

For each customer, the CLV is the sum of three estimates: 

• Customer Lifetime Value = Current Value  + Future Value + Referral Value
• Current Value = Sum of Revenue – Cost
• The Future Value is related to how long the customer will stay with the company, which can be estimated using the customer retention rate.
  • Average Customer Lifetime = 1/Churn Rate
• Another key factor for the Future Value is how much the customer spend will increase or decrease over time

For segments of customers, the CLV can be calculated using the formula below: 

Customer Lifetime Value = (Average Customer Margin per Year+ Average Referral Value per Year) * Average Customer Lifetime 

CLV through Customer Segmentation

Another popular approach for understanding CLV is through Customer Segmentation. 

There are many ways to segment the customer base, including demographic, geographic, revenue and behavioral segmentation.

Companies also start to use unsupervised learning models to segment their clients based on statistical similarity. Incorporating customer segmentation into the Customer Lifetime Value Model helps companies to better define and target their best customers. 

---

The Difference between Traditional CLV and Dynamic CLV  

Traditionally, marketers have been using the segment-based approach for calculating CLV as it is a quick and easy way to understand the general CLV for different segments. However, with the rapid development of data science, marketers have started to adopt the dynamic CLV calculation approach. This provides CLV estimates on an individual client level. 

The dynamic CLV calculation takes each client’s background information, product usage, historical complaint, and billing data as input, and generates a unique predicted customer lifetime value as output. The dynamic CLV prediction allows a much more granular understanding of each client’s overall value, which helps develop customized marketing and sales strategies.  

---

Key Components for Calculating Dynamic Customer Lifetime Values 

Based on the CLV formula(s) above, we can draw the following conclusions.

• The key components for calculating dynamic customer lifetime values are current value evaluation, future value prediction, and referral value calculation.
• For each of those components, we can find its corresponding metrics 
  • Current Value Evaluation <== Accurate Calculation of Revenue & Cost 
  • Future Value Prediction <== Customer Churn Probability & Upsell Probability 
  • Referral Value Prediction <==Customer Referral Probability 

We can build machine learning models to optimize the calculation of those metrics.  

For example, we can use customer information, product usage data, billing data, and complaint data to train ML models to predict the churn probability for each customer.

1. We will first label the historical data based on the churn definition.  
2. Then train supervised learning models, such as Logistics Regression, Random Forest, SVM, and XGBoost based on the target label to make the classification. 
3. Once we have trained and fine-tuned the models, we will be able to use the model trained to predict the churn probability for each individual client. 

Making Upsell Predictions with CLV

For upsell predictions, we will take numerous important factors into consideration:  the customers’ purchase behaviors, marketing campaigns, product promotions, etc. 

By tracking those changing factors for each customer, we will be able to derive the best prediction algorithm for upsell behaviors. We can also find the key contributing factors for upsell based on the modeling results.

Need Help Implementing Dynamic CLV?

In this blog post, we have discussed the advantages of Dynamic CLV Prediction as well as the methodology of generating the Dynamic CLV. While the methodology discussed above can be used as the general guideline for Dynamic CLV Prediction, the actual implementation tactics vary depends on the company’s system architecture, data quality, business metrics, etc.  

To maximize the effectiveness of the Dynamic CLV Calculation system, we recommend working with our advanced analytics team to create a customized solution. Building a full DCLV system requires the support of data scientists, data engineers, front-end and back-end developers. It is also computationally intensive to train machine learning models on real-time datasets for DCLV system. 

To achieve the best return on investment, a customized solution architecture design with quick prototype testing is recommended as the foundation for DCLV implementation.  

---