15 August 2016


Use Machine Learning to Drive Customer Retention

Use Machine Learning to Drive Customer Retention

All e-commerce businesses should measure customer retention but some don’t know where to start. Many want to use machine learning to drive customer retention but don’t do so because it appears to be complex.  As more companies adopt a subscription based strategy, measuring and maintaining a high customer retention rate is critical. Why? Simply put, it’s always cheaper to keep existing customer than it does to find a new one. The solution?

Let’s look at some back of the napkin calculations. If you lost 20% of your customers yearly, it would take a 25% increase in new customers just to get you even again. What if you could reduce that churn rate to 10%? Then it would only take an 11% increase in new customers to get you even again.

Sure, first term customers can generate good profits but long term customers are more valuable. A long term customer can be up or cross sold too and it’s easier to maintain those revenue streams. Losing customers not only affects your profits but makes you spend considerable time and effort to find a new ones. Any reduction in your churn rate translates to a higher retention rate and new opportunities for revenue.

So, how do you calculate customer retention and churn? That all depends on the company and industry they are in. These metrics vary from company to company but the best of the ones look at dozens of different metrics.

Classic Customer Churn

The classic way of calculating your customer retention is to first visualize the customers that have left. Any Business Intelligence (BI) tool (i.e. Tableau or Qlik) can help by building dashboards and reports. These dashboards are a great way to communicate what happened. But what if you want to find out in advance what will happen so you can prevent it? How do you determine what the right action is to prevent churn from happening?  How can you test complex customer behavior patterns?

Much of your customer data is “highly dimensional,” meaning that statistically significant relationships are typically buried deep inside the data.  You might try a solution only to find that what worked for one market area didn’t work in the others. Making things worse is that customer behavior tends to evolve as preferences change. The goal is to learn from your data quickly and take the right actions as close to real time as possible.

Learning the patterns of customer behavior from 6 months old data and taking actions based on that model completely misses the point. What good is it if you offer a customer a discount after they’ve left your company?

ML to Drive Customer Retention

Machine Learning has the ability to quickly and effectively analyze your customer data for those complex patterns.  Machine Learning and algorithms like Gradient Boost Trees or Generalized Linear Machines can understand highly dimensional data reliably. They can make sense of all the complex relations that drive your customer churn and help you strategize on how to stop it.

While the math behind these machine learning algorithms is complex – you often need 100’s of line of code – RapidMiner makes it simple.  The simple drag and drop interface of RapidMiner Studio hides the complex math and let’s you focus on the business problem at hand. This makes things more productive and lets you attack more business problems in the a fraction of the time it takes you to “code it.”

The following video shows you how to build a reliable, validated, and powerful customer retention model in minutes. This model is not just for one industry or company type, but can be optimized across many different industries. This is one step away from productive without have to write a “lick of code.”

Want to learn more? Check out our blog post on common challenges (and solutions) for building a customer churn model here.

Related Resources