Churn prediction in financial companies: How to get started



Satisfied customers are the foundation of most businesses. They generate a steady stream of income and are more open to upselling. At the same time, studies show that it costs five times as much to acquire a new customer as it does to retain an existing one. Therefore, it is important to focus on how you can retain existing customers. A key part of retaining customers is being able to predict which customers are on the verge of leaving - preferably before they even realize it themselves. It may seem like an almost impossible task at first glance.
May 11, 2023 twoday kapacity

Fortunately, you don't need a crystal ball to solve this; with the right data, business understanding, and machine learning, you can get started with churn prediction and predict who is at risk of leaving you in the near future.

 Although you want to sprinkle AI magic over your business, there are a number of considerations to make before embarking on churn prediction. Data work in the financial sector is a double-edged sword. On one hand, it offers fantastic opportunities; there is extremely detailed historical data on things like demographics, financial status, and past transactions.

 On the other hand, this large amount of data, comes with challenges; increased complexity can complicate access to data, as they may be stored in silos with different access levels. At the same time, the data is sensitive to personal information, so you need to keep your GDPR knowledge up to date. Fortunately, we will now guide you through the questions you need to consider, so that you can use machine learning magic best.

Who is it that we actually want to retain with churn prediction?

Rich data leads to rich opportunities. Therefore, you may quickly feel like starting with a huge model that can predict all types of churn. However, there are two reasons why this is not the best approach:

1. Machine learning works better with more specific definitions. 

In that way, the models can use more energy to capture deep patterns and interactions. This truly triggers machine learning's superhuman ability to understand large amounts of data.

2. More specific models are easier for the business to act on.

For example, if you create a model targeting small and medium-sized business customers, you can ensure that churn scores are presented to the right people in the business. This does not mean that you should lock yourself into one niche segment forever. As churn scoring is integrated into the business, you can build more models that hit different segments.

Fundamentally, there are two questions you need to answer to develop a successful churn prediction model. The first question is how to define churn. Often, customers can change their profile but still stay within the business, for example by changing their account type. In such cases, it may not be counted as churn, as no revenue is lost. A customer can also transfer a portion of their assets, such as all their bonds, to a competitor. Should that be considered churn, or should a churn threshold be set? These are the kinds of questions that ultimately determine how well the model can be incorporated into the business flow.

The second important question is who the target audience is. Not all customer groups are created equal. The behavior of a startup is markedly different from that of an elderly couple. This makes it more difficult for models to capture patterns if they are supposed to cover both groups. At the same time, the two groups are handled in different parts of the business, which can make the results less relevant. It is therefore incredibly important to ensure that a relevant and value-creating target group is targeted.

Both questions can be difficult to handle at first. But as long as good communication is ensured between the data scientist who develops the model and the business that is the end user, things cannot go completely wrong.

Which data are available for churn prediction?

A churn model is only as good as the data it's built on. Therefore, it's important to have an overview of the available data before starting the project. There are many different types of data that can be included, such as:

  • Historical data with customer transactions

    When and how has the customer used your product? Has it primarily been many small transactions or a few larger ones?

  • Customer master data

    If it's private customers, what is their gender, age, and possibly geography? What are their marital status? For business customers, it could be something like the company's size and industry.

  • Interactions with the customer

    Does the customer receive newsletters, and if so, are they being read? Has there been any form of correspondence? Text data can be a goldmine if you know the right tools. Read more about Natural Language Processing (NLP) here.

  • See a full checklist of the data you need for churn prediction here.

In addition to your own data, it's advisable to consider external data sources. It could be data from financial markets, demographic data from Danish statistics, or mentions of your brand on Twitter. Only your imagination – and skilled data scientists – set the limits!


How to get churn scoring implemented in the business?

Once you have sorted out definitions and data, you may face the most important question: How to get the most out of churn scoring? The most common use case is to target campaigns and offers. If you use a shotgun approach in your campaigns, you potentially waste a lot of resources on customers who had no intention of leaving you. 

Instead, you can use churn scoring to create a prioritized list of customers who are most likely to churn. There are steps you can take to make churn scoring even more valuable. For example, you can combine churn scoring with a customer segmentation model so that you can further tailor your campaigns to the customer type. 

Furthermore, you can add a so-called Customer Lifetime Value calculation (CLV), which provides a data-based estimate of the expected revenue from the customer. In this way, you can focus your energy on the most value-creating customers.

In addition to the purely operational benefits that churn scoring provides, it can also provide insights into the business. You may find that customers who receive newsletters are less likely to churn, or that people churn especially when they have just gotten married. You can also use special models to obtain information about specific customer cases, which can be beneficial for customer support.

 Churn scoring can truly become a valuable tool for retaining customers, gaining new insights, and creating more relevant campaigns. In this blog post, we have provided some general tips on what considerations to make before getting started. If you want to know more about your specific business, contact us at +45 26309001.

Learn more about twoday kapacitys Customer Churn & Retention Framework here.

Related posts