July 13, 2020

Using Data and Machine Learning to Improve ROI for Customer Retention

Bingqian Gao
Bingqian Gao Data Science Lead

Business challenge:

Customer Analytics
Customer Experience
Data Science
Machine Learning
Understanding customer insight

What does it mean to improve Return on Investment (ROI) for customer retention? 

Since there are two components in the equation: return (benefit) and investment (cost), if you want to improve ROI, there are 2 things you can do – increase the return, and reduce the investment.


In order to achieve better return with less investment, an organisation not only has to rely on the experience of Customer Experience professionals in creating and executing strategic plans for customer engagement programmes, but also should leverage the information hidden in data to validate and enhance the expertise, and inform decisions through concrete evidence. Organisations that are not embracing an analytics culture might be missing opportunities while their competitions are charging ahead.  

Having data-driven and AI-assisted decision-making at the centre of the Customer Experience and Customer Success functions in the business transforms the way they work – it arms them with the confidence and truths from the data. Finding out which customers are unhappy, will likely to churn and therefore should be the focus of the customer retention activities is no longer filled with “guess work” or as labour-intensive as before.   

Identify and Understand Dissatisfied Customers 

One of the common challenges for Chief Customer Officers (CCOs) and their Customer Experience (CX) team, is to identify and understand dissatisfied customers. This is an important step in getting to know what your customers want, and how your products or services are meeting or failing to meet those demands.  

Data analytics is an effective instrument to help CX professionals identify who the unhappy customers are and why. Start with Descriptive analytics to highlight where there are opportunities to improve customer experience. The CX team should be equipped with the foundational knowledge of the current landscape of the customer base. How fast is it growing? How do new customer acquisition and existing customer churn contribute to the growth rate? Who are the churning customers and what are their characteristics and churning behaviours? How does it differ for the different product or service offerings? How is it changing over time, and is there any noticeable trend or seasonality? Are there any geographical or market-specific patterns?  

Once you have a good grasp of the situation, we can dive in further with Diagnostic analysis. How can we better identify the happy and unhappy customers before they churn? Why do they churn? Does their churning behaviour correlate with any other behaviours exhibited throughout the customer life cycle? What tip them over? Can we segment the customer base according to their behaviours or key use case? 

Catch Early Signals – Predict Future Customer Behaviour 

Deep understanding of historical churn is essential, but it is not enough. More importantly, the CX professionals want to know how to predict customers from churning in the future. From the volumes of historical data, are there signals we can pick up that indicate customer health or satisfaction on an on-going basis?  

How to assess customer satisfaction and score Customer Health is a subject that is critical to the success of Customer Experience teams. Being able to quantify and monitor the customer health provides a solid foundation in allowing the business to create a consistent approach to improve customer experience and build loyalty. 

Three of the traditional approaches used widely are Net Promotor Score (NPS), sentiment analysis, and customer health scorecard (or Red-Amber-Green flag).  

Net Promotor Score is a singular and specific method to obtain an index as a proxy to gauge the customer’s overall satisfaction. With NPS, the data availability and quality is often an issue. In addition, particularly for enterprise products, WHO were asked to provide input WHEN may affect the answer and could result in ratings that may not be an accurate reflection of the organisation’s likelihood to renew a contract.  

For sentiment analysis, if the data source is survey data, same challenges will apply as the NPS. If the data is sourced from social media, some products are discussed more on than others (e.g. B2C vs B2B). Besides, the source systems (the chosen social media platforms) would already serve as a self-selection process which might impose innate biases in the data.  

A manually defined health score is a good starting point to combine different data inputs from various areas. However, the manual configuration of calculation rules means that the reliability of the score is heavily relying on the assumptions being accurate.  

Predictive analytics provides a more advanced and sophisticated way to model and predict customers’ future behaviour. By using Machine Learning techniques, the models can pick up signals from a rich selection of data sources, where it learns the pattern and relationship between the signals and churning behaviour from historical data. Then, the trained model can be applied to the latest data to predict the probability of a customer stop using or renewing your product or service offering in the future. This probability is essentially the predicted, quantified risk of a customer churning. You can set a threshold, and if the predicted probability is higher than the threshold, then the customers are flagged as needing investigation or intervention.  

Using Machine Learning algorithm to predict customer churn means that you are not relying on subjective judgement and manual configuration of what signals are the best at indicating customer health, and therefore have the strongest predictive power of future churn. The machine learns from historical data, which means the patterns it learned will be specific to your customer base (instead of a generic industry trend, though there might be similarities). And as customer behaviours change and data gets updated, the machine will also learn the updated pattern, which is pivotal especially during this unprecedented time of COVID-19. This means that when done correctly, machine learning models can often predict future churn behaviours more accurately compared to traditional approaches, which allows more targeted intervention and efficient use of resources in the Customer Experience department.  

From Insights to Interventions 

Having a thorough understanding of customer health, and an accurate prediction of future customer churn likelihood is only the beginning of the journey in improving customer retention. Knowledge and awareness will not improve ROI until the insights are put to use in designing interventions to improve customer experience.  

Knowing why customers are at risk will help determine the next steps and improve intervention efficacy. The workforce on the ground provides valuable insights from their direct interaction with customers. But Machine Learning algorithms can also help to steer the direction of investigation, especially when the insights from the ground is lacking or insufficient.  

Feature importance analysis allows deep dive into the model explainability, and is an effective technique in Drivers Analysis. It “unpacks” the machine learning model, and shows the contribution of the different factors in generating each prediction.  

Imagine on your customer retention dashboard, there is strategic customer who is at 90% risk of churning. How powerful would it be if you can click at the prediction to view the detailed contribution to this 90%, which makes you realise that the biggest reason for the high predicted risk is due to the fact that the customer has not had any marketing or sales engagement for the past 2 years? Hopefully, in your next team meeting, you know exactly who to get involved to define a plan of action. 

To summarise, extracting information from data using analytics and machine learning can help Customer Experience professionals identify and understand unhappy and dissatisfied customersflag at-risk customers who are likely to churn in the future, and design efficient intervention programmes that allows the team to have a more focused, targeted approach and achieve better intervention efficacy.   

If you found this article interesting, check-out ‘How to build an amazing customer experience using analytics‘.

Business challenge:

Customer Analytics
Customer Experience
Data Science
Machine Learning
Understanding customer insight
Bingqian Gao
Written by Bingqian Gao Data Science Lead

Bingqian believes in the power of Analytics and Data Science in uncovering insights and helping to better inform decision making. As a Senior Consultant and Data Science Lead at TrueCue, she enjoys finding solutions for challenges in data consolidation, modelling, visualisation and Advanced Analytics.

She leverages modern technology such as Alteryx, Tableau, DataRobot, and Microsoft Azure Machine Learning, and is one of the 17 Certified Alteryx Experts in the world. Outside of work, she enjoys a wide range of activities, from oil painting, poetry reading, scuba diving, to boxing and krav maga.

Case Study: Fighting COVID-19 with Data

Learn about how we’re helping the largest Sustainability and Transformation Partnership in the NHS, to use data analytics to improve their response to COVID-19 and save lives.

Download Case Study