How to Predict Customer Churn?

Simplicity is the Key

In the last blog post, I explained why the churn prediction is crucial to any business in any industry. If you have not read that yet and want to know why then go ahead and have a glance: Why is Churn Prediction important?

Now that you are up to speed, let's find out how to predict customer churn before they happen. A reminder that this blog post will be a little technical compared to the previous ones.

Churn Predictions 101 Prerequisites

Before showing how to predict, let me explain the building blocks of churn predictions. The intuition here is to observe and understand customer behaviors from the past and understand why a customer decides to stop purchasing your product and services.

There is a reason why one chooses to unsubscribe from your services or stops purchasing products from you. These reasons can sometimes be in the wild open and easy to find. However, sometimes these reasons are buried deep in the data that is not openly visible at the start with the naked eye, but these reasons are there and a machine learning model can peel those layers in the data and point the reasons out for you. Also, let's face it you can't go through all the data regarding all the customers manually, that would be preposterous.

You will need the Sales/Purchase data and Customer data for this predictive analytics.

Choices, Choices and More Choices!

It's needless to say, where there is data science involved there are countless choices presented at your feet. R is perfect for someone with statistical background, Python is the bread and butter for computer science people, SPSS is the true savior for social science people and IBM Watsons is the so-called rich kid problem no one talks about. These are only to name a few. My point is you can go on and on with the choices you have. I present to you my choice of platform. How can I put this delicately, this is what happens when minimalism meets the data-science. In other words, simplicity is the Key. The platform’s name is Enhencer.

Let’s Cut to the Chase

Enhencer is an online machine learning platform. Just go ahead and sign up for free, there is a free plan and trial period for premiums ones.

1. Okay, let’s log in to the platform and you will be greeted with a workspace and some demo projects. Just create a new project and ignore everything else for the moment. After you provide a name please choose the “Churn Prediction” from the list.
2. Now you will be presented with a screen to upload your dataset. Just choose the source that suits you the most and upload both the data.
3. On the right panel, you will find these two options. The first one is whether your churn is seasonal, if you don't know it then chances are it's Nonseasonal anyway.
4. Churn definition is the period of time after which you consider the customer as a churned customer. For instance, e-commerce might consider if a customer is not purchasing for 3 months then the customer might have churned. Just choose the period of time that is suitable for your data. Hit the “Explore” button and Enhencer will do its own magic.

Enhencer will:

• Clean the dataset and conduct Feature Enginnering automatically. This is big, because in a project like this, feature engineering takes the bulk of manual labor and it can takes days for this step alone.
• Train many Machine Learning Models on the data and choose the most accurate one Automatically.
• At the very end provide you with Customer Churn Predictions.

In a nutshell, all you have to do is upload the data and that's it, the rest of the steps will be done by Enhencer automatically. When I said I choose the simple method, I can bet that you did not take it for that simple. I guess I like to work smart not hard :).

The thing is it might be hard to believe a program can do all these automatically. But Enhencer has AI algorithms as its backbone that decides what to do and what not to do.

In this Case, the End Justifies the Means

So, does this level of simplicity actually help or hurt? One way to find out is to check the results. Enhencer will provide all the results and output in a dashboard like this.

Let me break down the most important elements from the dashboard for you. The data we uploaded had a total of 2474 customers; among them, 32% churned in the past.

Enhencer also segments the customers based on their past behavior and their likelihood to churn in the future. Segment 1 contains the customers with the highest probability to churn, meaning these 19 are the customer most risky. There is a 77.8% chance that these customers will churn in the next three months.

Enhencer provides the behavior of these customers as well. These 19 customers who are very, very likely to churn have a few common behaviors. In the past, these 19 customers bought at least one product from more than one product category. They have been inactive for at least 55 days. They gave a product rating of below 1 and are more than 27 years old.

This is a very powerful insight. This means you can reach out to these 19 customers with some marketing campaigns that are very much personalized based on their behavior. In other words, offer them something they can’t refuse, and you will have a very high chance to retain them and prevent them from becoming churn.

Similarly, if you reach out to customers from, let’s say, top 5 segments and organize marketing campaigns personalized based on their behavior, this will bring down churn significantly in the long run.

All you have to do now is to press the “Export Customers” button and the churn predictions for all your customers will be downloaded. You can connect to your own platform directly using the “Get API Code” button as well. So just suit yourself.

Realize how fast, in matter of minutes, we ended up with these absolutely insightful outputs. This would have taken days in other methods or platforms. I guess in this case the ends do justify the means and the simplicity did pay off.

The Aftermath

So you have predicted who will churn in near future, now what? Well no2 comes the hard part. At this point you know which customers will churn along with their segment behavior. All you have to do is act on those two information. Go reach out to those customers with high probabilities and provide them offers, promotions or whatever suits your company according to their segment behavior. This way you marketing campaigns will be more effective then ever. This will help you retain more churn likely customers before they actually churn. In other words, Enhencer is the fortune cookie you wish you had earlier.