Churn is one of the most important metrics by which an operator is measured. In a saturated telecom market such as the U.S., it becomes particularly important to keep customers because it is so much more expensive to acquire new ones. But churn-related analytics models are also tricky because they must catch or predict relatively rare occurrences: a model could predict that a user in a given circumstance will not churn, and at many telecom operators more than 97% of it the time, it would be right. It’s that 1-3% of users that do churn that operators want to identify and focus their retention marketing efforts on.
Big data analytics is increasingly being used to get to a holy grail of a 360-degree view of the customer and his or her experience, hoping that better information and intelligence can be used to improve customer satisfaction and provide a path to proactive customer care that can address the reasons for churn before a customer actually takes the action to cut off service.
Mobile marketing company Globyssaid that in a recent instance, it was able to get a 300% improvement in customer churn prediction accuracy and a 400% improvement in retention marketing success compared to traditional methods, by using rapid, behavior-based customer churn analytics in its Amplero solution.
Optimizing retention marketing is still largely a manual process that can take weeks or months to determine what has worked in order to establish the most effective offers and redistribute them to the customers that an operator wants to target, according to Lara Albert, VP of global marketing for Amplero. The data on which models are based also tends to be injected infrequently — on a monthly basis, for example, rather than on a more frequent or near-real-time basis.
Dr. Olly Downs, chief scientist and CTO at Globys, said that churn drivers can include network quality and quality of experience, a customer’s device, and the service experience he or she has. He said Amplero is differentiating its approach by presenting a longitudinal, timeline-based view of the customer that includes every event in his or her customer experience. Having that detailed, long-term view means being able to garner insight on overall trends that lead to churn and can be prevented. But Downs also said that Amplero’s approach in near-real-time takes a huge lag out of subscriber churn analytics: the monthly cycle, which is the basis upon which much data for churn models is available. Trends and flags, he said, are typically built upon a two-cycle or 60-day window — which entirely misses, say, the customer who tries out a new service or device, has an unsatisfactory experience, and leaves before their trial period is up.
“When you have a data trend that looks at a month, or month-over-month only, and updates scores monthly, they don’t change,” Downs said. That monthly view, he added, slows down the pace at which customers can be flagged and the carrier’s reaction time for retention marketing. Running Amplero’s model side by side with traditional churn models, he said, resulted in both the dramatic increase in churn prediction accuracy as well as even better lift for the retention marketing. Downs said that the “sweet spot” for retention marketing leaves a carrier with between 2.5-3.5 weeks to reach out to potential churners, so that they have time to do the marketing that could retain them.
Traditional models, Albert added, may only flag churners after they’ve already left or only provide enough lead time for a carrier to have one shot at retention, compared to a model that scores based on user interactions and behavior changes up to the last hour and can support a series of supportive interactions that may address a customer’s issues.
For more on big data and analytics in telecom, download RCR’s recent special report on Implications and Applications of Real-Time and Fast Data Analytics, and view the webinar.
