Knowing Your Customers – Part 1: Tackling Customer Churn

Article
By
Anne Constance
October 29, 2021 8 minute read

 

  • It is a well-known fact that customer acquisition costs about 5 times more than customer retention.[1]
  • According to the inventor of the Net Promoter Score, Frederick Reichheld, a 5% increase in customer retention could result in an increase of profit by 25–95%.[2]
  • Existing customers are likely to spend more money than new customers.[3]

What do all these customer insights add up to, and why are they important?

Churn (or attrition) can be passive, involuntary, or even “delinquent” – mainly caused by tools’ mechanical failure (e.g., credit card issues, including expiration dates, expense limits, and security updates) – and is therefore quite easy to resolve. However, customer churn can otherwise be active and intentional, and it is this second category that we’re interested in today. Although churn is inevitable to some extent (about 50% of customers naturally churn every 5 years)[4], it is critical for a business to be able to identify the early warning signs of customer churn, as this has a direct impact on financial metrics driving value (monthly recurring revenue, long-term value, customer acquisition cost) and, consequently, overall revenue.

Customers can churn at different stages and should therefore be treated differently at each stage. Think of it as the process of looking for a new employer/employee:

In the short term: first impression is everything.
You are on the lookout for the slightest opportunity to give up, to change your mind. However, you know that you cannot grasp the full context at a since glance – it will take time for you to be sure.

Somehow, it all comes down to your experience. Did you have a good time, and does your gut tell you to stick with your original intuition? If so, you will go for it and agree to make an investment. Likewise, when first using a product or service you have paid for, the onboarding experience needs to be frictionless, close to perfection.

In the middle term: was the initial promise kept?
To continue being satisfied with your choice, there needs to be constant effort to build trust and keep things as delightful as it was during the first stage. Bugs and glitches are to be expected, but communication is key to resolve any problems encountered.

In the long run: nuance is key.
At this stage, you need to be reminded of the reasons you made a particular choice and stop worrying about potentially missing out on better options. You will need signs that you are one of a kind and still valued: through upgrades, exclusive rewards, incentives, and more.

Identifying early signs of churn:
Now that we have established what churn is and why addressing it is critical, we need to identify how to tackle the problem. As in any human relationship, it is always easier to fix a problem in its early stages than later, when things get more complex and polyfactorial.

This is where Machine Learning (ML) comes into play. Statistical models and extensive computational power help predict customers’ likelihood to churn within a predefined period: advanced algorithms do so by identifying patterns, raising alerts upon spotting anomalies or “outliers” – unusual data not conforming to expected behavior – and helping decision-makers act fast. For instance, when generating statistics on a population’s height, if one person is listed as being 4.87 meters tall, this data would be considered an anomaly (or in ML terms, an outlier). The same applies when attempting to identify fraud (e.g., abnormal purchases, unauthorized access, abnormalities in insurance claims & payments), structural defects (e.g., flaws in car production, energy leakage), medical problems (e.g., cardiac rhythm monitoring), fake news, and so on.

Let’s dive into how analytics can help detect customers at the risk of churn. With ML’s ability to automatically learn a set of rules from data, it can effectively enable customer churn prediction in the following ways:

Identifying what features and/or combinations of features (i.e., consumer characteristics) are crucial to determining whether a customer will churn or not:

This can be done by learning from the averaged predicted mistakes compared to the actual results (through a linear regression model). For example: What is going to best explain whether a customer will renew his gym subscription next year? Is it the number of hours he spends working out every day? Is it the number of children he has? Is it the fact that he does not smoke? Is it the fact that he is in his 30s? Is it because he is a non-smoker AND has no children? Although easy to understand, this model may not always be very accurate.

Learning from past observed decisions to predict future decisions with tree-based models (e.g., Random Forest, XGboots): 

This relies on algorithms successively splitting customers into two groups at each step according to the “most accurately (or “purely”) splitting explanation”. For instance, if someone stops buying a certain yogurt, the computer will learn if the first (and most accurate) split is to be man vs woman, under vs over 35 y.o., high vs low risk of diabetes, and so on. It will then continue with this process until it finds the best series of decisions. These models are very accurate in their predictions but do not provide any information about drawbacks in customer experiences.

Mimicking the way the human brain works (hence “Artificial Neural Networks”) while modelling interactions between elements that would otherwise go undetected, with the machine learning techniques mentioned above:

Deep learning models are highly complex and are especially useful for analyzing large amounts of data (aka big data), for example, recognizing faces, translating texts, winning at Chess or Go, driving autonomously, or even identifying cancer cells. However, here again, these algorithms do not provide information on why customers decide to churn, only what churning customers have in common.

Adopting a People-Centric Strategy
For effective and sustainable churn reduction, these short/mid-term ML analyses need to be complemented with longer-term investigations on the why, the reasons causing the churn: to identify the root of the problem, resolve it, and increase customer loyalty.

Indeed, it is critical to remember that customer onboarding and their journeys should be people-centric. A service/product is going to be used by humans and should be tailored to their pain points, not simply be outputs of a company’s functions or technical innovations. Today, uberization, social media, and digitalization have raised the bar of consumers’ expectations. They are highly informed, and compare and benchmark every offering before making a choice. In addition, they value experience just as highly as products or services, which has led companies across the B2B and B2C sectors to compete against one another to deliver unparalleled customer experiences. The customer journey of someone who wants to buy a bag from a physical store needs to be as simple as for the person wanting to buy a car digitally, the payment method for a real estate purchase is expected to be as quick and secure as that for a train ticket, and so on and so forth.

As customers are purchasing a specific outcome from your company, your product or service is a tool to help them get there. Therefore, their experience should reflect the problem you are solving for them instead of merely showcasing your product/service’s features, and communicate clearly what the brand is committed to doing for its customers across all touchpoints (offers, products, services, CRM, loyalty, merchandising, and so on).

Last but not the least, your AI/ML insights and efforts should be closely tied to your understanding of customers’ pain points (e.g., through social listening, customer sentiment analysis, etc.), as well as your customer personas to effectively build long-term customer relations, ensure a high level of customer engagement, and evoke strong brand loyalty.

Read Part 2 of the Knowing Your Customers series here.

Bibliography:

1. Gallo, Amy. “The Value of Keeping the Right Customers.” Harvard Business Review, October 29, 2014.
https://hbr.org/2014/10/the-value-of-keeping-the-right-customers.

2. Li, Susan. “Modelling Customer Churn When Churns Are Not Explicitly Observed, with R.” Medium, April 23, 2018.
https://towardsdatascience.com/modelling-customer-churn-when-churns-are-not-explicitly-observed-with-r-a768a1c919d5.

3. Habif, Stephanie. “Making Sense of Customers Who Cancel.” Medium, March 9, 2017.
https://medium.com/behavior-design/making-sense-of-customers-who-cancel-901e04ba3bd0.

4. Toxboe, Anders. “Making the Fogg Behavior Model Actionable.” UI Patterns, April 8, 2019.
https://ui-patterns.com/blog/making-the-fogg-behavior-model-actionable.

5. Dancho, Matt. “Deep Learning with Keras to Predict Customer Churn.” Posit AI Blog, January 11, 2018.
https://blogs.rstudio.com/ai/posts/2018-01-11-keras-customer-churn/.

6. Kazmi, Rashid. “Unlocking Behavioural Secrets to Overcome Churn Extremes.” Medium, July 23, 2022.
https://towardsdatascience.com/unlocking-behavioural-secrets-to-overcome-churn-extremes-9c12a7ccc875.

7. Loyalty myth #8: It costs five times more to acquire a … – ipsos, n.d.
https://www.ipsos.com/sites/default/files/publication/2003-08/Ipsos_Loyalty_Myth_8_Excerpt.pdf.

8. Stillwagon, Amanda. “Did You Know: A 5% Increase in Retention Increases Profits by up to 95%.” Small Business Trends, October 2, 2014.
https://smallbiztrends.com/2014/09/increase-in-customer-retention-increases-profits.html.

9. Taiwo. “Repeat Customers Spend More – Tracksend – SMS Marketing Automation Platform.” Tracksend, September 7, 2022.
https://tracksend.co/repeat-customers-spend-more/.

10. Afshar, Vala. “50 Important Customer Experience Stats for Business Leaders.” HuffPost, December 7, 2017.
https://www.huffpost.com/entry/50-important-customer-exp_b_8295772.