Tapping into Big Data for Predictive Customer Churn Analysis

A 3G user receives an SMS from the customer care center one weekend morning. The message is about phone packages—get a free phone with certain amount of credit. In fact, the user’s service subscription is about to expire, and they may be thinking of buying a new mobile phone and switching to another operator. Just hours ago, they browsed B2C websites over the 3G connection to check the price of mobile phones mentioned in the message and compare all available bundles. The message arrives just in time with an attractive offer. The user also remembers that 3G access in their apartment is slow and that a new phone could be a handy alternative way of using the internet. The user makes a short detour to the nearest telecom business hall.

Operators are feeling the pressure of churn because customers are eyeing up attractive bundles offered by competitors. If the customer doesn’t have a good reason to stay, the competition will give them a good reason to leave. How can operators attract new customers and retain old ones in the fast-growing big-data era?

The Importance of a Customer Churn Prediction Model

Mobile operators focus their attention on bringing in new customers and exploring new business opportunities. Mature technologies and an increasingly saturated telecom market have led to fierce competition between operators. People are free to choose more cost-effective packages or better quality service. Customer churn is no longer rare for operators. It is less expensive to retain a current customer than to find a new one. Statistically speaking, acquiring new customers can cost up to five times more than satisfying and retaining existing customers. In a mobile market with fewer and fewer incremental customers, lower churn means less cost and less revenue loss. Therefore, mobile operators have to be concerned about customer churn. Invigorating an existing customer base has become a hot topic. The most common ways of reducing customer churn are stored charges, free charges, and free phone packages. If operators can accurately predict in advance which users may leave, they are more likely to take early action to prevent churn and minimize losses.

Telecom managers seek to understand which customers may leave and when. A customer churn prediction model can help managers predict customer behavior by analyzing historical and current data, extracting key data for decision making, and discovering hidden relationships and patterns. Churn prediction models have been a hot research field in recent years. Operators have spent a lot of time and effort creating and improving models and have achieved results. T-Mobile US has integrated multiple IT systems into a large data application. By combining mass customer history data and analyzing customer transactions and interactions, T-Mobile could extract the behavior of the customer before they switched to another operator. This helped T-Mobile accurately predict its customer churn. In the first quarter of 2011, T-Mobile halved its churn rate in the United States.

Deficiencies of Existing Churn-Prediction Models

An accurate churn-prediction model depends largely on the completeness, quantity, and quality of the available data. Factors such as brand, bandwidth, terminal, service, consumption behavior, tariff, convenience, change of workplace, and user experience can all be causes of customer churn. Operators can never get all the information about customers but can make assumptions based on the available information. In this case, the decisions an operator may make might not be optimal (or even completely wrong). Therefore, an operator needs to make an effort to collect and integrate new data about customers from emerging contact points. This allows an operator to gain insight into the wants, preferences, and decision-making processes of their customers.

Certain events, behaviors, and environments can be identified as having bearing on customers prior to a churn event, and these factors can be predicted. For example, an operator might find that a customer’s data traffic significantly drops every month in the lead up to the customer leaving. The number of monthly outgoing calls may decrease, and there may be records that the customer has made complaints to the service center. In this case, the customer is a high risk to leave. At present, most telecom companies are using customer, network, and service data extracted from business analyses, CRM, billing, and network management systems to establish a customer churn model. Such data include customer age, gender, occupation, type of terminal, call records, traffic, complaints, home region, geographical location, online time, date of churn, and payment information. Although analyzing this data allows an operator to predict customer churn to some extent, the real reason a customer leaves can never be fully known every time. A customer may have changed simply because the competitor has a better quality network.

Market research firm Synovate surveyed mobile phone users in more than 8000 cities in Ukraine, Russia, India, Indonesia and Argentina and found that on average 48% of users believed that network quality was the primary factor influencing their choice of operator. That is to say, a user is most likely to give up a network if they find that the call is not clear, network coverage is not wide enough, web pages are hard to open, or social media is slow to update. Therefore, the churn prediction model includes user experience as a criterion for accurately identifying these users. This can greatly increase customer retention.

Improving the Customer Churn Prediction Model with CDR and IPDR

The call detail record (CDR) and internet protocol detail record (IPDR) are important data sources for quantifying and analyzing user experience and behavior. However, because of technical complexity and lack of related technical standards, CDR and IPDR data have not yet been widely used by operators for customer churn prediction.

CDR and IPDR data can be used in various applications as a new type of data source. From the data, operators can determine to whom calls are being made, how often calls are being made, where the user is located, how good the signal is, how long the service is used, what web pages the user has browsed, what mobile internet applications the user has used, how often applications are used, and how good  application performance is. These new data sources are very beneficial for information analysis. When new data is input into the churn-prediction model, user experience features are sorted using deep data mining, and an operator can understand relationships between factors prior to a churn event. With such knowledge, an operator can then determine whether network quality was an issue leading up to the loss of the customer and determine how the network can be optimized to retain customers in the future.

ZTE is an expert at managing user experience and has unique insight into CDR and IPDR data analysis. ZTE’s UniCare technical service solutions provide operators with complete, custom-made, end-to-end CEA user experience guarantee and OC operational consulting services. These solutions help operators rapidly improve network performance and address customer churn issues.

Conclusion