Have you ever imagined predicting the loss of your customers? And, knowing that you are about to lose them, be able to carry out customized retention actions? Some years ago, this seemed like an impractical task due to the restriction of customer information and technologies to carry out the prediction with a good degree of assertiveness. But today, with the growing adoption of concepts of Big Data & Analytics in organizations, the use of advanced analysis techniques to identify opportunities throughout the customer lifecycle has been an increasingly common practice. As presented in the graphic below, we highlight the opportunities to use these techniques to leverage efforts related to the customer lifecycle.
Learn More: about Big Data & Analytics
Big Data refers to a set of very large and/or complex data. Through well-developed techniques that are already widely marketed via cloud services, Big Data allows us to work on this large volume and variety of data with high processing speed.
In turn, Analytics techniques represent a work of intelligence and analysis of these large volumes of data, structured or not. Interpreted by high-performance software, which allows this information to be crossed quickly, the techniques assist and support decision-making.
The methods of Machine learning, or machine learning, have been enough
used to predict customer loss. This technique uses algorithms that process large volumes of data to identify patterns of interest and thus make predictions. In the case of retention, methods of Machine learning can be used to understand customer behavior just before Churn (abandonment). Identifying behaviors similar to the established pattern allows you to focus efforts on specific retention actions.
Overall, a good forecast of customer outflows - combined with a targeted commercial approach - can bring great results for companies. Focused retention actions, such as price renegotiation or benefit offers, can extend the lives of clients and offset the costs of the actions taken.
How to predict the loss of customers?
Before applying any analysis method, it is important to reflect on what causes a customer to stop being one. The origin of the abandonment may be directly related to the experience lived with the company. In other words, the Churn may occur and, consequently, be understood through the analysis of negative expectations and possible frustrations of the customer with the product or service provided since the beginning of their relationship. It is through understanding the causes that the initial objectives of the development of predictive modeling are established. That is, based on known results and behaviors, a model will be developed that can be used to predict future results.
Once the causes that lead to the loss of a customer are understood, the information that potentially makes it possible to predict the Churn. This information may vary from company to company and depends on the relationship established with the customer - contract, subscription, one-time purchase, and others. In general, the factors that predict loss can be grouped into 5 categories: customer profile, cost of change, marketing actions, competition, and customer satisfaction.
Learn more: Categories of information with the potential to predict the Churn
There are several examples of how the customer profile can affect the Churn. Some examples verified in studies show that higher-income clients are less likely to Churn — perhaps because they are less sensitive to price changes, as is the rural population. People with lower education, in turn, have higher rates of Churn. It is important to carry out complementary analyses to identify specific profile patterns for the analyzed case and not to use only the examples mentioned here or in other studies.
The lower the cost for the customer to switch to a competitor, the more prone they are to do so. These costs can be psychological, which include inertia, brand search, familiarity, and sense of relationship with the current company, or physical, which are those actually related to the value or effort of the change.
Loyalty program discounts and rewards relate to lower rates of Churn, just as businesses with personalized products have fewer Churn compared to those with the ideal of “one size fits all”.
Competition can take place within a category when, for example, it occurs between different pay TV operators, or between categories, such as what occurs in pay TV operators with the growth of online streaming services.
It's intuitive that happier customers are less likely to Churn, and this theory is proven in several studies. However, a generic measure of customer satisfaction may not be so related to Churn. It is important to have specific indicators about customer satisfaction, for example, if the product meets expectations, if it meets the customer's needs, if the price pleases, etc. As such information is difficult to obtain, previous purchases and usage behavior are often used as substitutes. In addition, you can include Help Desk calls and complaints as customer satisfaction assessment data.
A series of analyses must be carried out to identify factors that are truly relevant for predicting the phenomenon of Churn and subsequent submission of the information collected to the methods of Machine learning. Several techniques can be used, such as decision trees, regressions, and neural networks. These algorithms seek to identify customer behavior patterns before leaving and are thus able to estimate their propensity to Churn.
After identifying the clients with the highest exit risk, they are normally prioritized for the development of retention actions according to criteria such as the probability of leaving and also their value to the company. The use of control groups, which do not receive any type of retention action, is also common practice to serve as a basis for comparison and measure the performance of predictive models.
Forecast case of Churn at a financial services firm
The use of Machine learning For the forecast of Churn of clients can be illustrated by a recent case in which Visagio had the opportunity to support a company in the financial services sector in identifying the clients most likely to cancel its service. The company faced a growing retention challenge mainly due to the reduction of entry barriers previously established by regulatory bodies in the sector, allowing more and more competitors to seek space in the market.
Before the start of predictive modeling, an in-depth analysis of the company's retention landscape was carried out, identifying the profile of the customers that leave, the abandonment rates, and the performance of the ongoing retention actions. Over 200 potential predictor variables related to customer profile and behavior before Churn were identified by mapping the interactions and possible frustrations that the customer could have with the company. By studying the correlation of each of these variables with customer abandonment and their importance in explaining the phenomenon, it was identified that 10 of them were in fact relevant to the prediction of the event. The application of various techniques of Machine learning the information collected, together with the process of constructing and adjusting the predictive model, resulted in a prediction with performance approximately 7 times higher than the technique used to select clients in preventive retention actions.
Machine learning It's not a silver bullet against Churn
The market has already proven the use of advanced analysis techniques, such as those of Machine learning, as powerful tools to support organizations with customer retention challenges. More and more companies are investing in areas dedicated to the development of analyses of this nature and are reaping great results from customer information.
The methods of Machine learning to predict the departure of customers should not, however, be considered as the solution to this problem. Although predictive models provide valuable information to direct retention actions, customer loyalty will actually be maintained by good experiences and a real competitive advantage seen over other competitors.
About the authors
Gilberto Cordeiro is a partner at Visagio, a specialist in projects focusing on Analytics having experience in the health, financial, mining and agribusiness sectors.
Michael Schardosim is a Visagio consultant, a project specialist focusing on Analytics having experience in the financial sector.
