How to develop Customer Retention Strategies?

Customer retention is very important because acquiring a new customer is far more expensive than keeping an existing one. This is even more true if the market operating is saturated or saturating as in mobile telecommunication industry. Retention is important to most businesses because the cost of acquiring new customers is much greater than the cost of keeping good relationship with current customers.

Do you know why your customers are leaving and who they are?

Can you develop effective retention strategies without knowing this vital information? Will it be possible to do so? The single most important thing in customer retention is to understand your customers well enough: customers' expectations, satisfaction, demographic & geographic & psychographic customer tendencies, etc. If you understand more about customers and know more about which customer groups are defecting to rival providers, more effective retention strategies can be developed. This is the key to the successful retention marketing!

Customer churn detection methods

For obvious reasons, the most important customer retention strategy is to identify customers who are likely to churn (potentially to rival providers). Once they are identified, customer retention programs can be developed and actions can be taken. The following analytic techniques (customer churn analysis) can be used to identify customer groups having high defection risk;

• Churn trend analysis: Analyze trends in customer segments. With Sales Trend Analysis, customer segments with increasing churn trends can be identified.
• Customer churn profiling: Develop profiles of churn risk groups based on demographic, geographic, psychographic attributes and service usage patterns.
• Customer churn scoring and segmentation: Build predictive models that can predict likelihood of defection and perform segmentation based on defection scoring.

Customer churn trend analysis

Customer churn trends can be analyzed with time-series analytic tools described in sales trend analysis. It can identify trends of customer segments in various ways: geographic, demographic, psychographic factors, and so on. It uses time-series predictive modeling to identify trends in sales amounts, customer numbers, etc. Sales trend analysis is a powerful tool for detecting churn trends. The following figure shows sales trend by customer segments. The sky-blue bars represent recent sales figures. The red bars indicate next three term forecast. "Trend(3)" uses three recent figures for modeling. So it indicates short-term trends. "Trend(6)" uses six recent figures to show mid-term trends. For more, please read Trend Analysis.

Customer churn profiling

Profiling churns can reveal customer segments and the reasons why customers are leaving your service. Customer information, when all combined, can be very large. It may consist of dozens of fields. Analyzing data with many variables with conventional tools is a challenge. Using StarProbe hotspot analysis tools, profiles of high-risk groups can be accurately identified instantaneously. The following figure shows an example of hotspot output;

Customer churn predictive modeling and churn scoring

Customer churn scoring is a very appealing approach provided that accurate scores can be computed. Generally speaking, customer defection is relatively a low ratio event. Developing predictive models on low event data is extremely difficult and therefore models tend to be inherently un-reliable. Fortunately, neural network provides a way to identify customer groups that have shown to have high churn ratios in high accuracy. The following figures show an example of customer churn scoring produced by a customer churn predictive model. Note that the left (or next) chart shows proportional distribution.

Reds are customers who have churned. They are concentrated at high churn score ranges. Customers scored high but not churned have high probability of churning in the near future. In the histograms, they are represented in green parts. Preventive actions should be taken to those current customers with high scores. In the histograms, customers scored over "0.2" are the ones who may need preventive measures! Note that they are very tiny groups. Cost effective measures can be implemented.

Pitfalls of classification predictive modeling techniques Intuitively, classification predictive modeling is a very appealing technique as it predicts outcomes in a very simplistic manner: churn or no churn. Modeling tools supporting classification include decision trees, SVM, neural network, etc. One major limitation of classification techniques lies with the fact that defection rate is generally too low to make classification techniques work reasonably well. Developing classification models with skewed data is especially difficult. Decision trees develop predictive models through segmenting populations into smaller groups repeatedly. It uses the dominant value (or most frequent value) of each segment as the predicted value for the segment. Dominant values are the values represented by over 50% segment population (in case of binary decisions as in churn predictions). Low-rate nature of defection will lead to decision nodes with non-convincing decisions. This type of models will have very low accuracy in predicting on-risk customers correctly, therefore missing most defectors! Mobile phone customer churn example Mobile telecommunication industry has high churn rates. Industrial churn prediction on mobile phone telecom showed that best classification predictive models built had about or less than 30% accuracy for identifying defectors correctly. This means that classification models will miss over 70% defectors. It is noted that amongst 30% customers, many of them will leave whatever actions telco takes. So most customers who are likely to defect will be gone! As a rule, classification techniques should not be used as the sole method for identifying potential defectors. Rather it should be used as a supplementary method to the ones described in this page!

How to implement advanced churn detection techniques?

Advanced churn detection techniques can be implemented using Rule-based Modeling Environment (RME) very easily. RME incorporates rule-based control with advanced predictive modeling, providing most robust churn detection environment. Once churn detection models are developed, they can be applied to customer database on a regular basis, say, weekly, monthly, etc. This will identify customers who have potential for defection but have not been contacted for retention purposes in recent times. Preventive actions can be followed for customers identified as churn risk!