Churn prediction and management

Home - Article - Churn prediction and management

You’ve done it. You built a killer marketing funnel, found the right channel, and have great Customer Acquisition Costs. It’s time for your business to take off. Except for one problem: churn.

It’s as if you’ve built the house of your dreams but the wood is full of termites. Churn is that hidden problem that can turn even amazing success in customer acquisition into ultimate failure.

That said, churn is a manageable problem on a portfolio level.

The Hard Facts about Churn

Retaining a customer is 5-6x cheaper than attracting a new one. This means you’re going to get far more ROI from focusing on churn prevention rather than wasting money pouring new customers into a broken funnel.

It’s no wonder that a modest increase of customer retention by 5% can translate into anything from a 25% to a 95% increase in profits.

While fixing churn can even help the top of your funnel when you utilize word of mouth or other growth channels driven by existing customers.

If all that sounds great, what do you need to do to start addressing your churn?

That’s the problem we’re going to address here: not being able to properly manage customer retention (churn). But what lies in-between knowing about the hypotheticals and getting actionable information to tackle churn?

Why the Old Approach Is Failing

Let’s start with the typical approach. This relies on hiring expensive experts and consultants to evaluate our portfolios on an ongoing basis.

But what lies in-between knowing about real business-case and converting it into action and hence value? Well, the typical approach so far has been to rely on expertise and specialised human resources that evaluate our portfolio on an ongoing basis. While this is a valid solution for some companies, it generally can’t keep up with the highly competitive and fast-paced business environment which demands automation, predictability, and flexibility. 

The efficiency and cost reductions that come with automating churn prediction and management are becoming more and more essential.

This automation-driven approach is a part of the greater transformation that is Revolution 4.0, the fourth industrial revolution represented by the integration of IoT (Internet of Things) into every aspect of manufacturing and logistics.

All that isn’t to say that there’s no use for experts and that they can be replaced entirely with this new generation of automated systems. The real key is combining these two valuable resources to use each where it can bring the most value: domain experts for strategic thinking and analysis while machine learning and data science derives actionable insights from enormous data sets.

Practical Steps for Tackling Churn

Here’s a list of general, experience-driven steps for tackling churn (and similar ones, see the picture above for reference):

No alt text provided for this image

Here’s a list of general, experience-driven steps for tackling this problem (and similar ones, see the picture above for reference):

  1. Understand your pain-point and hence your business case.
  2. Find the right data for this problem, for example: socio-demographic, product features, financial relationship with the customer, etc. Ensure the data is in the proper historical time-horizon.
  3. Exploratory analysis of your portfolio: review your customers through the magnifying glass of data science and business intelligence, and understand patterns, potential segments, etc.
  4. Use advanced predictive analytics to model your pain-point: automate the process of proactively identifying customers that are about to churn. Align the predictive power of the model with the necessary inference level so that you can understand what drives these events (meaning you start understanding your customers better).
  5. Integrate the predictive model into your CRM to provide: a) highly efficient, automated and manageable customers churn scoring system; b) targeted campaign based on improved multilayer understanding of the customer.
  6. Track your data and activities to ensure you improve your bottom-line over time (see the section below for a what-if scenario).

What-if (TELCO industry example)

  • Thanks to the scoring system we were able to score all clients and sort them by their probability to churn. Our interest lies in those with 70-100% probability to leave the company. These people attribute for about 10% of our portfolio;
  • Thanks to visual analytcs and KPIs we know who these customers are: predominantly using Fiber Optics; Tenure < 1 year; usually responsible for less than 500 currency units of profit;
  • Campaign: we target 107 people of which 75 are Churners.


No alt text provided for this image

Don’t Stop There

Still, this combination of experts and machine learning doesn’t mean the problem of churn is solved forever. You’ll still need to regularly re-examine your approach.

We strongly recommend you monitor your results and models. It’s essential to fine-tune them and every-so-often and even redeveloped over time. The business conjecture is constantly changing as well as we are so we need to evolve.

Truthfully yours,


With this series our aim is to increase data science coverage and to make data-driven decisions an integral part of more companies around the Globe.


More to read