Skip to content

Machine Learning to the rescue of your most valuable metric

Share on twitter
Share on linkedin
Share on email
Share on whatsapp
Machine Learning

In my previous post: Learn how to predict customer churn with Data Mining and Machine LearningI talked about Machine Learning and how it could help insurance companies. In that post, I not only talked about Machine Learning, but also about the importance of customers in insurance companies, the cost of losing a customer and the importance of customer retention.

At that time, we were delving deeper into predictive models of customer leakage and the benefits these bring to companies. Today, in my new post I want to talk about a new term Customer Lifetime Value and how some Machine Learning techniques can help us predict it.

Customer Lifetime Value, or CLV, is defined as an estimate of the profit or loss resulting from the company's business relationship with a customer, taking into account the length of time the relationship with that customer will be maintained.

The easiest way to calculate CLV is to add up the revenue the company obtains from a customer and subtract the cost to the company of acquiring and managing it. To calculate the CLV we can apply several formulas, one of the simplest is the following:

CLVSimple = TP * RA * VC


  • TP: average purchase of a customer.
  • AR: frequency of purchase.
  • VC: average duration of the relationship.

Who could use this measure within an insurance company?

This value allows marketers to analyze different aspects of customers, and make appropriate decisions to improve the company's customer portfolio.

In addition, it is important to note that this measure not only allows you to know who your most valuable customers are, but also provides long-term value.

How does this value help to improve the customer base?

These are some of the issues that insurers have on the table when talking about improving their customer base, and about which CLV has a lot to say:

  • Does the Marketing channel invest enough in attracting new customers?
  • When you have already acquired a new customer, is he the best type of customer, or is he a short-lived customer who is difficult to keep?
  • It is possible to adjust spending on campaigns that help build loyalty and maintain a good relationship with the company's current customers.
  • How to change the way you see your customer, so that they become an important part of the company you have to invest in to keep them.

Once we know what CLV is, we can say that insurance companies could benefit from knowing this value for each of their customers and what in addition, they could use it to resolve issues such as those just raised.

Can insurance companies increase their current CLV?

Here are some examples of actions that companies should focus on to increase their Customer Lifetime Value

  • Build customer loyalty. Or, in other words, make your portfolio more and more old customers, whose relationship with the company is growing.
  • Increase the number of products your customers purchase.
  • Improve customer service, so that customers feel satisfied with the service and products purchased.

And, can we predict the future value of the customer? Can we predict which customers are more profitable for the company and which are not?

The CLV value can be calculated in two ways: historically, for specific time periods, or on the other hand, it can be calculated in a predictive way. Both calculations are equally valid, but each for a different purpose.

  • Historical methods look at past data and make a judgment about the value of customers, based solely on past transactions with the company, without any attempt to predict what those customers will do next.
  • Predictive methods aim to model the purchasing behavior of customers in order to infer what their future actions will be.

There are many who claim that predicting the value of CLV is the most powerful way to understand not only what a customer is worth to the company, but also how their value will change over time.

Focusing on predictive methods

When modeling purchasing behavior to predict CLV, the business context on which you want to model is very important. The business context of insurance companies is defined by: being a contractual environment and whose purchase opportunities are not continuous.

In these methods, is where machine learning techniques come into play, because when we talk about modeling, we are talking about managing the information that the company records of the activity of its customers, and use it to train the so-called predictive models. And before starting to train these models, it is necessary to evaluate which are the ones that bring us the most benefits for our current objective.

When it comes to evaluating probabilistic models, in particular the models used to predict CLV tend to focus on the same three constant parameters, which characterize customer behavior:

  • The period during which a client maintains its relationship with the company.
  • Number of purchases a customer will make during a given period of time.
  • Expense of each future customer transaction.

Among the best known probabilistic models, the main characteristic of value that each of them brings us when modelling is

  • The Pareto / NBD model allows you to calculate the expected number of purchases in a forecast period for each customer.
  • The Gamma-Gamma model, being an evolution of the previous model, allows assigning a value to each of these future purchases.
  • Another option is the combined use of both models for the forecast of the CLV of each client; for this purpose, the values of the expectations thrown by each model are multiplied and their results are more complete.

What else can we learn about the value of CLV?

CLV is considered one of the best metrics to predict a company's future customer behavior. There are many applications in which this value is used and here I leave you with a selection of the most interesting ones:

  • Some scientific articles bet on using Customer Lifetime Value to model customer leakage patterns, as the European Journal of Operational Research tells us in this article.
  • Improve customer service performance
  • Improve the identification of the target audience for each of the marketing campaigns.
  • Make personalized e-mail campaigns.
  • Remind customers when a product or service is going to expire soon.
  • Apply cross-selling techniques.

Share the article

Share on twitter
Share on linkedin
Share on email
Share on whatsapp

A new generation of technological services and products for our customers