Академический Документы
Профессиональный Документы
Культура Документы
A case study into the viral effect of churn and product adoption in the telecommunications industry
CONCLUSIONS PAPER
Featuring:
Carlos Pinheiro, PhD, Head of the Analytics Lab, Telemar Norte Leste S.A. (Oi) Ken King, Director of Telecommunications and Media Convergence, SAS
Table of Contents
Fundamentals of Customer Link Analysis . . . . . . . . . . . . . . . . . . . . . . 2 Customer Link Analysis in Action: Reducing Churn . . . . . . . . . . . . . . 2 Of Those Who Churn, What Differentiates Leaders and Followers?. . . 4 How Do the Churners Compare to the Remainers? . . . . . . . . . . . . . . . 4 Considering Communities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Putting Customer Link Analysis to Work in Churn Propensity Models . 5 Customer Link Analysis in Action: Boosting Product Adoption. . . . . 6 Proving the Concept with Another Communication Service Provider. . 6 Not All Leaders Are Created Alike . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 About the Presenters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 For More Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Churn is an inescapable reality in the telecom world. Customers come and go, swayed by the latest disappointments or the latest deals. This churn is costly. It is far more expensive to acquire a customer than to satisfy and retain an existing one. With the pervasiveness of social media, dissatisfied customers also have a big soap box to bash the brand. Its just better to keep them happy. Analytics has proven its value in identifying customers at risk for churn and helping marketers understand how to retain them. In developed markets, most major service providers have adopted churn propensity models that have succeeded in reducing churn for post-paid subscribers from about 3 percent a few years ago to about 1 to 1.5 percent today. We might not be able to attribute 100 percent of that decrease to analytic intelligence, but certainly the models have delivered dramatic improvements. How would you like to improve analytic model results even more say, make them up to 30 percent more accurate? You can, if you acknowledge that consumer purchase behavior and loyalty are increasingly influenced by friends and acquaintances. Imagine the power of understanding social relationships and using that information to better target the customers who count the most. What if you could identify community leaders the ones with the most clout so you could stop them from influencing 10 of their followers to churn? On the buy side, what if you could identify the leaders who influence 250 percent more product adoption than others? If you knew who those leaders were, you could target them wisely and extend the impact of your marketing and retention efforts. In a SAS-sponsored webinar, Carlos Pinheiro, PhD, of Oi, Latin Americas secondlargest telecommunications provider, shared research that dramatizes just how powerful social networks and their leaders can be and why marketers cannot afford to overlook this insight. Pinheiros propensity models for both churn and purchase behavior go far beyond demographics and transaction history. These models factor in a host of social network analysis metrics, such as weighted detail about the number, closeness and strength of connections among customers. When applied to marketing questions, this type of analysis is called customer link analysis.
In a superconnected world, consumer purchase behavior and loyalty are increasingly influenced by friends and acquaintances. Imagine the power of understanding social relationships and using that information to better target the customers who count the most.
What it reveals
The number of ties to another node in the network, as in seven degrees of separation. The relative weights of links and nodes to adjacent nodes, based on multiple factors. The weights of the nodes adjacent to nodes adjacent to a given node n. The number of important nodes to which a node points. The number of important nodes that point to a given node. The degree to which an individual is near all other individuals in a network, directly or indirectly reflecting the potential for contact or influence among them. The percentage of possible time spent by other nodes to node n, similar to the page ranking in a Google search. The number of shortest paths in which node n partakes, which clarifies how a particular node controls information flow into the network. A measure of the likelihood that two associates of a node are associates themselves. A higher clustering coefficient indicates a greater chance of clique connection.
Pinheiro walked through the process of a customer link analysis project for a European telecom provider. The process started by aggregating call data for two to three earlier months to build the social network, identify communities, and understand relationships within the communities. At a designated point in time Time 0 Pinheiro flagged customers who decided to churn and then analyzed what happened with related customers. In this particular example, the analysis built the network and calculated communities based on call records from September, October and November; flagged customers who decided to leave the company; and then looked at January and February data (Time 1 and Time 2) to see what happened with other customers in the churners communities. Did their departures have an effect on others? The data-gathering process looked at more than who is calling whom. Nodes and links were weighted by using data about the type of call (pre-paid or post-paid, incoming or outgoing, voice or text message, etc.), as well as call frequency and duration, average revenue per user (ARPU) and other factors. In the first time period under study, Time 1, approximately 81,000 customers churned. Of those churners, about 49,000 were active in the network; that is, they had frequent contact with others. These customers in turn were related to 624,000 more customers, suggesting a wide pool of potential influence. Of the nearly 50,000 active users who churned, 128 could be considered leaders, said Pinheiro, because after they churned, so did 134 of their connections. Those 128 leaders were connected to more than 1,100 of the companys subscribers, suggesting that the leaders could influence the behavior of about 11 percent of the customers to whom they are related. If you consider that these leaders can cause big damage in your customer base if they decide to leave, these are customers you definitely need to track, said Pinheiro.
With customer link analytics, you can quickly identify not only customers at risk for churn, but also the connections between customers and identify leaders, followers and other members of their social communities.
Figure 1. Among subscribers who churned, the leaders influenced the behavior of about 11 percent of their connections.
Figure 2. Compared to the average churner, leaders are more valuable and well-connected, and influence more information flow.
How do followers those who churn after someone else in their network did compare to average churners overall? Pinheiros study found that followers represented 29 percent less ARPU but had 200 percent more distinct connections (115 percent more strong connections) and 127 percent more connections to important customers. In short, although the followers were less valuable to the company than churners in general, they were very well-connected to distinct and important customers and had strong links to their connections. That insight demonstrates the importance of understanding subscribers not just based on their value as individuals but as nodes in a social network.
So although churners were well-connected to distinct customers, they showed distinctly less loyalty. Almost half of the churners had come from competitor B, so they know how to use portability in telecommunications and are more likely to use this capability in the future.
Considering Communities
Using a community detection capability within customer link analysis, Pinheiro identified 814,000 distinct communities. When looking at churn among all 28 million members in the social network under study, Pinheiro found the average rate of churn was 0.83 percent in Time 2 and 0.74 percent in Time 3. Those figures look pretty heartening. But when we looked at churn as related to communities, we saw that that the average rate of churn was 2.7 percent in T2 and 2.72 percent in T3, said Pinheiro. So we can consider that in some communities that have had churn in a previous time period, we will see more likelihood of churn in later times, even among those who would have been remainers. In this particular case, the difference was more than three times the average rate. This is food for thought, once again showing the power of connections and the importance of understanding these connections to know which customers to target for retention efforts. How do communities with churn compare to communities overall? As you might expect, churner communities represented 18 percent more ARPU, had 44 percent more members, and had 64 percent more distinct connections (42 percent more strong connections). Of note is the fact that these communities had 67 percent more off-net members, suggesting that connections who are competitors customers can exert an influence.
Customer link analytics is the perfect tool to help you understand the viral effect in social networks and trigger better marketing and sales campaigns.
Ken King Director of Telecommunications and Media Convergence, SAS
Figure 3. Social connections can be highly predictive indicators of customers future buying behavior.
With customer link analytics, you can quickly visualize social networks between customers yours and your competitors and uncover leaders, followers and other members within social communities. Social network metrics from this analysis can enhance existing segmentation models, enabling you to discover how best to target influencers.
Figure 4. As with churn, social influence on product adoption is very evident in communities.
Conclusion
These figures can definitely prove how important it is to understand the viral effect on any sort of business effect, said Pinheiro. In this case, we looked at churn or product adoption. If you understand who are the leaders and who are the followers, you can establish better and more focused marketing and sales campaigns or loyalty campaigns to increase the effectiveness of your business actions. Communications service providers should be putting aside the concept that customers are only viewed by their individual attributes, and look at how they relate to each other, said Ken King, Director of Telecommunications and Media Convergence at SAS. These studies required pulling a lot of data together and bringing a lot of complex analysis to the question, but in the end the research was able to demonstrate some exceptionally strong benefit to understanding social connections.
Ken King is Director of Telecommunications and Media Convergence at SAS, and a global expert on the use of advanced business analytics to solve complex business problems for the communications industry. He consults with SAS clients and industry thought leaders on topics such as churn management, social network analysis, product bundling, offer optimization, forecasting and operational improvements. SAS solutions are deployed in almost every large telecommunications service provider. King has more than 25 years of global telecom industry experience in designing, deploying and supporting voice and data networks. He has extensive experience in migration of communication networks from circuit to packet switching. His background includes roles in sales, marketing, product management, alliances, engineering, testing, training and technical support. King has a BS in computer and information science from the University of Florida and an MBA in product innovation management from North Carolina State University. He has dual Australian and American citizenships.
About SAS
SAS is the leader in business analytics software and services, and the largest independent vendor in the business intelligence market. Through innovative solutions, SAS helps customers at more than 55,000 sites improve performance and deliver value by making better decisions faster. Since 1976, SAS has been giving customers around the world THE POWER TO KNOW . For more information on SAS Business Analytics software and services, visit sas.com.