Churn Analysis
Predict who is getting ready to leave your bank and increase your retention.
Financial institutions are among those with the highest customer acquisition costs in the industry. Estimates show that acquiring new customers can cost five times more than retaining existing ones.
In today´s highly competitive market, where information is readily available and switching costs are ever-decreasing, customers in any industry barely hesitate to move their business elsewhere, should they find more acceptable alternatives to meet their needs.
Opening a new account at a bank has become a more simplified process, sometimes fully digital. The open banking regulations have also paved the way for a more straightforward process for transferring information and accounts to other banks. Both of these realities make it easier for your customers to try out different options and will provide a chance for your competitors to become your customers’ preferred bank in the future.
As such, it is in the best interest of any company to keep a close watch on its customers to monitor any signs of potential churning down the line. Additionally, because customer acquisition typically has a higher cost than customer retention, maintaining clients has become more and more critical to a company’s success.
What is Churn Analysis?
Significant headways in the Business Intelligence field have brought forward a great number of tools for knowledge discovery and predictive analytics purposes. Churn analysis focuses exactly on that.
Customer churn is the rate at which customers stop purchasing more of your products or, ultimately, stop using your services, by closing the account. It is often used as an important KPI in Sales or Marketing departments, which can be defined on different timeframes and compared to benchmark results in the industry.
Churn analysis is a method to measure these rates. It will tell you the percentage of customers that churn compared to those that keep doing business with your bank. Using these numbers, you will be able to identify some trends and promote business initiatives to prevent future customer churn.
However, since it is a lagging KPI, not a leading one, you won’t be able to act on it until it’s too late for those customers. To more effectively prevent Churn, you will want to go one step further and conduct a more advanced analysis: predicting the customer churn rate for particular clients.
Through the timely identification of the customers most likely to leave your business, you can focus your efforts to understand their possible complaints and provide them with a better customer experience, that will ultimately contribute to retaining them.
Defining churn for your business
To start measuring your churn rate, we first need to define what churning is for your context.
A churned client can be defined in multiple ways, for example:
A client who closed their account with the bank;
A client who stopped using the bank services, such as credit or debit cards, but hasn’t closed their account;
A client who stopped receiving their salary in their main account;
A client that has moved their credit to another bank.
Then, as in the previous chapter on Lead Scoring, you will need to identify key features that could help predict your client churning down the line, for example, their age, their income, their expenses how long have they been your customer, how many products they have, if they use your bank to receive their salary, if they have loans on other financial institutions, the bank’s share of wallet on these loans, etc.
You can then define business rules based on these features and establish business initiatives to contact the customers when any of those rules apply, such as when the client stops receiving the salary in your bank’s account.
Churn analysis with Machine Learning
Luckily with Machine learning, you won’t need to define business rules manually anymore. Just as we saw in the previous chapter, you can use the historic data of customer features coupled with records of previously churned customers to train supervised machine learning models that will predict if the client is likely to churn.
Typically you will want to know sometime in advance before clients churn, so a common way to define the target variable is if the customer churned between one or two months after the date when you calculate the features.
One can go even further and predict between six to eight months for example, but beware the results will be worse the further down the line you are trying to predict. Another approach is to build several churn models: one long-term, one medium-term, and another short-term. This way you can define different business initiatives based on both the likelihood of your client churning and how urgent you need to act.
| Churn model | Score | Business Action |
|---|---|---|
| 1 to 2 months | High | Improve conditions on existing products, regardless of long-term commitment. |
| Medium | Offer special conditions for new products the customer might need. | |
| Low | Do nothing. | |
| 2 to 6 months | High | Improve conditions on existing products. Ask for some form of long-term commitment such as a salary account. |
| Medium | Understand if there's a problem with the account manager. | |
| Low | Do nothing. | |
| 6 to 12 months | High | Understand if there's a problem with the account manager. |
| Medium | Offer special conditions for new products the customer might need. | |
| Low | Do nothing. |
Here’s an example of an operational initiative following the deployment of 3 churn prediction models, long-term, medium-term, and short-term, where churn is the client closing the account.
Preventing Churn
Retaining all customers regardless of their value to your business may not be a wise decision. It's important to consider the cost and value they bring to the company before deciding to retain them. By considering the value they bring, you can establish a minimum threshold for business actions to ensure retention efforts are worthwhile.
Our Experience
We’ve built churn models for a client in the banking sector. In this case, the definition of churn was clients who had not accessed the digital channels for more than 90 days. We approached this problem by creating 3 models with varying time windows:
Short-term: 0 to 30 days
Medium-term: 30 to 60 days
Long-term: 60 to 90 days
This enabled the bank to perform specific actions to get their feedback and reactivate said clients, increasing the retention rate on the bank’s digital channels.
Read about more churn analysis use cases here.
Want to learn more?
Miguel Cabrita
Senior Data Scientist, Co-founder
Miguel has helped various companies in banking and finance implement lead scoring and AI solutions.
Having a strong technological background and understanding of business processes and the banking industry helps him detect specific needs and offer the necessary AI solutions for each of them.
