Unlocking the potential of big data to stop telco customer churn

Zeng Zhi and Yang Yi
15 Jul 2013

Sponsored article

One of your friends, a current 3G network user, receives a short message from the customer care center when he is shopping on a weekend morning. It’s a telecom package offer: prepaid one-off tariff plus a gift mobile phone. In fact, the current subscription of your friend is about to expire and he is thinking of switching to another network and buying a new mobile phone. Just hours before, he has been browsing B2C websites through the 3G connection to check on the price of this mobile phone mentioned in that message and comparing all available packages online. The message comes just in time with a tempting offer! Better still, this guy also finds that the problems of slow 3G services, micro-blog and Internet browsing in his new apartment are solved without his consciousness. With a light heart, your friend steps into the nearest telecom business hall.

Does this story sound like a fairy tale? How can a mobile operator successfully keep his customer from churn? Perhaps you also have realized that in an era when big data see rapid development, we’ll soon get used to such retention stories.

Importance of customer churn prediction model
In an era of rapid business growth, most of the mobile operator's attention is focused on the development of new customers and exploration in new areas, that is, emphasis on the increment. Today, the mature technology and saturated market have led to fierce competition between telecom operators and users are free to choose the more cost-effective packages or better quality of service, that is, they can always be ready to migrate to another network, hence the customer churn. According to the statistics of Customer Satisfaction Manual of the American Marketing Association, the cost of retaining a customer is 1/5 of that of acquiring a new customer. For a mobile communications market with less and less incremental customers, lower churn rate means less cost used to reduce revenue loss. Therefore, mobile operators have to be concerned about the customer churn and hence the increasingly hot topic of “revitalizing the stock customers”. Commonly used customer retention measures are prepaid packages, tariff for free or handset for free. If operators have the ability to more accurately predict in advance which users may leave, they are more likely to take earlier actions to prevent the churn and thus minimize the losses.

Telecom managers want to understand when and which customers may leave. The churn prediction model, by analyzing historical and current data, extracting key data for decision support, and uncovering hidden relationships and patterns, can help the managers predict behaviors that may occur in the future. The churn prediction model is a hot research field in recent years. Operators have spent a lot of time and effort to create and improve the model and have achieved certain results. For example, T-Mobile, a mobile operator, integrated large data applications in multiple IT systems and by combining huge history data of customers and analyzing customer transactions and interactions, it extracted pre-churn characteristics of lost customers and therefore more accurately predicted the churn rate in the first quarter of 2011, and managed to have the churn rate fall by half in the US region.

Defects of current customer churn prediction model
An accurate churn prediction model depends largely on the comprehensiveness, quantity and quality of the available data. A number of factors such as brand, bandwidth, terminal, service, consumption behavior, tariff, convenience, change of work place and user experience can be the incentive to churn, however, operators can never get to know all customer information but make assumptions based on limited available information, that is, to infer a holistic picture with a glimpse. In this case, even if not completely wrong, the picture can hardly get near to optimal. Therefore, a telecom operator needs to make efforts to collect and integrate more new data sources of customer information from emerging contact points so as to get better in-depth insight into individual desires, preferences and decision-making processes of their customers.

It’s found in the study of actual churn events and churn environment that a great relevance exists in many ways of the lost customers. That is to say, predictive factors can be identified in regular behaviors or conditions before a churn event happens. For example, You may find that before some customer stops a service, his service consumption significantly drops by every month, his monthly call initiation attempt percentage falls and he may have repeatedly called to complain about the service. Now if another customer also meets these criteria, he represents a quite high churn risk too. At present, most of the telecom companies are making extensive use of the customer, service and network data extracted from the business analysis, CRM, billing, and network management systems for the customer churn analysis modeling. These data include age, gender, occupation, type of terminal, call records, traffic, complaints, home region, location, survival time, date of churn and payment information. By examining these data, operators expect to draw meaningful predictions. Although such a way allows operators to predict customer churn to some extent, they can in no way learn about the real incentive of customer migrations and thus cannot make prompt moves to keep customers from switching to other networks, for instance, whether the customer is simply upgrading to a better quality network.

Related content

No Comments Yet! Be the first to share what you think!
This website uses cookies
This provides customers with a personalized experience and increases the efficiency of visiting the site, allowing us to provide the most efficient service. By using the website and accepting the terms of the policy, you consent to the use of cookies in accordance with the terms of this policy.