Growth Hacking

How To Fight Churn: An AI Approach

Pinterest LinkedIn Tumblr

Why is there so much theory on conversion optimization in relation to churn prevention?… 🤨

Smart business people know that business growth isn’t just about attracting new customers. Sure, bringing in new people is exciting, and seeing the numbers grow makes you feel good, but it’s actually not the answer to profit.

Customer retention is where it’s at. Because whilst growth occurs when new customers, if too many customers leave it counteracts the growth and can even lead to contraction. Did you know that the cost of acquiring a new customer can be higher than that of retaining a customer by as much as 700%, and increasing customer retention rates by a mere 5% could increase profits by 25% to 95%?

One factor is customer lifetime value (CLV). Over a long period, one customer who comes back time and time again, always increasing their spend or adding to their basket, will be worth far more than a few customers who pop in once for the cheapest service on offer. 

Attracting customers takes a lot of effort – and therefore costs money. We’re not saying that you don’t have to put effort into looking after your existing customers, of course you do – but that effort manifests rewards many times greater.

What Is Churn?

The metric we use to measure this is churn.

It looks at how successful you have been at acquiring and retaining customers. What we classify as a ‘lost customer’ has a few nuances:

In some cases the customer actively ceases their relationship with you, such as by terminating an account. This is called absolute churn. In other cases, they don’t change their status ‘officially’ but do stop engaging with you – presumed churn.

The time period to measure by also varies. For a newspaper, a customer could be classified as lapsed after only a couple of weeks of not buying, whereas for an airline it could be many months, and a car company would be looking at years.

How the churn happens is another consideration. It’s called reactive churn when it’s something specific, such as bad customer experience causing someone to cease dealing with a business, and prospective churn when nothing specific has happened, but there’s a slow moving away from engagement.

To manage churn you need to be able to identify the triggers and address them. This is where an AI driven approach comes in.

How AI Can Help To Fight Churn

1. Understanding The Relationship Between Negative Triggers And Their Effects

Using historical data, businesses can study the link between negative customer experiences and customer response, and address them. If it’s known that this trigger is likely to lead to a customer leaving, it’s important to remedy it. This can include internal triggers, such as customer engagement, and external, like price cuts, although the latter are difficult to deal with.

2. Understanding Fading Behaviors

Historical data can also be used to predict prospective or slow driven attrition. By mapping existing customers with historical customers who have fallen away, it’s possible to predict who are at risk of lapsing. They might be exhibiting behaviors such as reducing frequency of purchase, not using loyalty cards, or cutting down on engagement.

3. Identifying High Risk Clusters

Using a decision tree model allows to segment the customer population into clusters linked together by specific behavioral characteristics. It’s then possible to see which have a high churn risk.

For example, you might discover that customers who join less than a year ago and spend less than $25 a month are more likely to leave. You then know that they are high risk and can treat them appropriately.


Let’s take a look at a fictional company – a data analytics software for building and sharing real-time business dashboards that can be created and use by multiple users.

The figure here  uses a technique called Behavioral Cohorts to show the relationship between a behavior and churn.

🧐 How to read: Each pool of customers, along with a behavioral metric, gets organized into a specific cohorts depending on their measurement in that metric. Usually ten cohorts are used, so the first cohort contains the bottom 10%, the second the next 10%, and so on.

As the bottom of the behavioral cohort plots is always set to zero churn rate, so the distance of the points from the bottom of the cohort plot can be used to compare relative churn rates. So if one point is half as far from the bottom of the plot as another that means the churn rate in that cohort is one half the other.

It also reveals that the churn rate on the cohort with the lowest active monthly users per month is around 8 times greater 😲 than churn in the cohort with the highest number of active users. 

The next figure show the license utilization metric calculated by dividing the number of active users by the number of seats the user has purchased – a seat means the number of users allowed.

This shows that license utilization is a very effective metric for fighting churn: the lowest cohort in license utilization has average utilization just above zero, and the highest cohort has license utilization around 1.0, and the lowest cohort has a churn rate that is nearly seven times that of the highest cohort. This makes it more effective for distinguishing churn risks than solely looking at active users.

You want to do this analysis also for your business? Get an account for RetentionX.


Beware Of The False Positives

One issue with churn management is the high prevalence of false positives in churn prediction models.

How can you determine, if a customers is just late with his next purchase or is already churned? 🤔

They have significant detrimental impact on programs, driving costs up, for no reason. Essentially, you would be spending money trying to retain customers who were never really at risk of leaving.

There are ways round this.

One step is increasing the number of data points, both in terms of types of data (e.g. enrich transactional data with customer experience data, marketing campaigns data etc.) and volume of data (e.g. number of months of data being used).

You can also exploit model heterogeneity by selecting different models that may be most effective for specific predictions (e.g., SVM, Neural Nets, RBM, Collaborative Filtering) and bringing them together to optimize overall performance.


What Data Is Required For Churn Management?

As is true in most cases, the more data you have, the better. Try to get a 360 degree view of every way that your customer engages with your business, giving you a fully holistic picture.

👉 Transactional data, such as spend across categories, purchase frequency and SKUs bought, as well as product attributes.

👉 Customer demographics like age, gender, lifestyle and location, and their experience with you on a loyalty program.

👉 Pricing, rates, timelines and schedules, and engagement with you through services and history.

👉 Marketing, campaigns and offers and how they responded.

👉 Macroeconomic factors such as interest rates, unemployment rates, economic growth, and seasonal effects.


The All-Important Question Of ROI

Of course, it is not enough to just identify the customers that are at a high risk of churning. You have to do something about it, and make it worthwhile.

The investment in addressing the attrition risk at an individual customer level has to reap rewards. AI and machine learning can help here, and offer three key steps to treat likely churners in the most effective way.

👍 Determine Customer Attractiveness: Not all customers are equal. The first thing you have to do is work out how important – which usual means how valuable in dollars – each individual customer is to you so you can spend your money on those who have the highest lifetime value for you. Modeling can use spend, revenue, wallet size, basket size, frequency and other patterns to be used to define attractiveness.

👍 Identify Customers’ Sensitivity To Treatment Inventory: Next it is important determine what are the most effective treatment strategies for each customer. You need to consider the type of campaign or intervention they would best respond to, and the types of communication channels that are most effective in each individual situation.

👍 Test And Learn Feedback Loop: Even knowing which customers you want to go after and which strategies might work best, you still need to be testing and learning. It’s important to be able to respond quickly, and learn from the strategies you place and the responses they gain. A nimble and agile organization can continually learn and hone their strategies using a fast feedback loop.

None of this is easy to do, but it is important. Thankfully there is software out there that does the job.

Pioneering market leader RetentionX translates your data into clear actions. By using AI-driven data analysis in an easy to use tool, it allows you to make business changing decisions driven by effective machine learning. As the most sophisticated data science engine out there it uses modeling, machine learning, and artificial intelligence to analyze data, allow you to benchmark stats, and deliver insights that you can act on.

 

 

Sources:
Shamli Prakash, „AI 101: Understanding Customer Churn Management“
Carl Gold, „Fighting Churn With Data“

 

Data Science Enthusiast, growing revenues with the power of data – Founder & CEO of RetentionX, the leading Software for Decision Intelligence.

Write A Comment