This paper outlines an approach to investigate and manage customer churn to enhance client retention and loyalty. Customer churn, or attrition, measures the number of customers who discontinue a service or stop buying products in a given time period. Measuring churn, understanding the underlying reasons and being able to anticipate and manage risks associated to customer churn are key areas for continuous increase in business value.
Businesses should first understand churn through comprehensive analysis to quantify its impact. Descriptive analysis will give initial insight, potentially identify specific groups of customers who are likely to churn. This can be followed by predictive churn modeling which will help in understanding the key factors of client attrition, identify the clients most at risk of leaving, and provide targeted insights on specific retention actions that should be implemented.
Business should identify focus areas for customer churn and then quantify the associated business impact – both for short and long term. It starts with something as simple as agreeing on criteria to define what constitutes a churn, types of churn, priority focus area, inclusions and exclusions etc. For example, the next consideration is whether to focus on hard and soft churn or both. Hard churn refers to a defined event that signifies churn, for example, the closure of an account. However, this approach may be too simplistic, thereby requiring soft churn to be considered. A customer may be defined as having soft churned if he or she has not transacted with the business for a certain period. The length of time varies depending upon the nature of the industry and often on the customer’s initial behavior (namely, transactional frequency). For example, a shorter period would be used if focusing on supermarket shopping (as we shop and eat every week) compared to booking a holiday.
Consideration should be given as to whether to focus on a specific customer group. The measurement may be limited to customers with the most valuable product, for example, home loans for banks, premium card holders, members of premium loyalty cards bucket or to another area such as the busiest time of year, Diwali, Christmas for seasonal businesses. No transaction by the premium or loyalty card holders during Diwali or other festive seasons might indicate first signs of losing that customer. Once the degree of churn is known, it needs to be expressed as a financial consequence to the business. So if the churn rate is x%, business will lose INR Y million.
To assess the financial impact depends on the value or profitability by product, service, and channel and/or customer segment. A number of assumptions based on customer knowledge can be made. For example, if the average tenure of a customer is 5 years, the expected future revenue can be built into the assessment.
To understand more about the type of customers that are churning involves profiling by comparing customers who have churned as compared to customers that have not churned. This will use all available information, including service history – behavioural, demographics, channel, value, recency, frequency, usage etc. The behavioral profile identifies typical customer patterns of behavior and interactions with the business prior to churning. Integrate any qualitative market research that is available to help build an understanding of the complete customer landscape for the churn events.
Predictive Modeling process starts with the creation of the analytical data sets that includes base and derived variables. Insights from descriptive analytics also serve as input for model building. A variety of modeling techniques can be applied on top of this data set once some degree of variable selection or feature engineering is done. Model assessment in terms of lift and accuracy can be used to determine which modeling technique is likely to yield better results. Classification table is one of the ways to assess model effectiveness. Cost of customer communications, expected revenue/profit can be factored in to arrive at a final decision regarding the set of customers to be considered for marketing action.
The objective is to optimize the number of accurate predictions that will appear as indicated by a smiley and thunderbolt in the chart above.
Key findings and analyst recommendations based on the analysis provide the deployment strategies for implementing customer retention and customer loyalty marketing initiatives to minimize and control the future impact of customer churn. Examples of the type of insight that one can expect include the best time to contact a customer based on a moment of truth and a defined demographic profile. The combination of likelihood to churn in conjunction with variables provides a powerful basis for marketing action.
A year’s worth of data covering the consumer annual cycle can provide some level of effective analysis. More historical data will add to the richness of the analysis. There are two key aspects to measure: to know whether the model is working as anticipated and to know whether the communications in place are effective or not.
Control group 1’s retention rate is compared with the customers’ retention rate within the retention program. Both groups have a high likelihood of churning and this comparison measures the effectiveness of the communications. Control group 1’s retention rate is compared with control group 2’s retention rate. Neither of these groups has received a communication and the expected result is that a higher number of customers are lost from control group 1 than control group 2.
Churn analysis is the first essential step towards implementing effective customer retention and customer loyalty programs. More specifically, it will:
- Establish the cost of customer churn to your business and provide justification for appropriate investment in customer retention and customer loyalty initiatives
- Target retention efforts on high-value customers with a high risk of churn
- Focus retention efforts on the areas over which your business has most control
Combat churn by providing the data backbone for developing predictive modeling for more proactive customer management.
Nishit Mittal is working with the Consulting-econometrics and analytics research team at Hansa Cequity and is based out of Mumbai ofﬁce. Nishit is a data science enthusiast and is trained in data science and business analytics from IIM Bangalore and LSE. He has worked as an economist with RBS and UBS and as a consultant with NCAER before joining Hansa Cequity. He can be reached at email@example.com
Chetan Bhangdia is Executive Vice President with Hansa Cequity. He oversees provisioning of solutions and build-out of capabilities in the areas of data management, analytics, technology, campaign management and customer relationship centers. He can be reached at firstname.lastname@example.org
Try deep learning using MATLAB