Mobile Social Network Analysis – The Next Big Thing in Telecom Analytics

TELECOMS

By Dr. Jay B. Simha, Chief Technology Officer, ABIBA Systems

WAR in Telecom

Wallet share, Acquisition and Retention, commonly called as WAR, form the basis of major revenue generation activities in telecom.

Providing only one service for a fixed rate is not profitable unless the usage trend is increased across the subscriber base. It is easier to predict marginal utility of such commodity services and it is more likely to reduce when the competition offers a similar service at a cheaper rate.

This requires having a set of services and products, which can be cross sold to the subscriber base. Further a subset of the customer base may be prone to increased usage of the same service, which requires identification of such subscribers and executing an upsell campaign.

Churn and customer value are critical to telecom. If a customer spends (ARPU) $50/month and an operator has 5M subscribers, then 0.5% churn is equivalent to a dent of $1.25M/month, which results in a cumulative loss year on year. Annual churn rates in the prepaid segment average between a significant 50 to70 per cent. Lowering this churn percentage has a large effect on the bottom line. 

Even a small reduction in churn can mean big savings – the cost of retaining a client is estimated to be only one-fifth of acquiring one. And these consumers could ultimately help decrease the churn within their own social circles, amounting to even more potential savings.

Mobile SNA

Mobile Social Network Analytics can help operators to reduce churn by studying the social behaviour of their subscribers. Mobile Network Analytics is hidden cousin of social network analytics. There are two types of MSNA:

• Based on the voice connectivity network within an operator’s customer base
• Social network on internet through mobile/smart phones

Voice based MSNA is a big data opportunity, hitherto was difficult to tackle. With the innovation and adoption of Hadoop based technologies, it is becoming a reality to add crowd/social information from the network in predicting the subscriber behaviour.

Mobile usage data available in CDRs will contain wealth of information. Mobile social networks (unlike other social communities), is mostly virtual community latent in the data. Each person can participate in multiple communities, creating a handful of opportunities for innovation for growth.

Table 1: Frequency of calls to groups (in percent) 

Reference: Sadie Plant, 2001. “On the mobile: The effect of mobile telephones in social and individual life”

It is a common sense that every individual has his own personal, professional and social networks. This creates a different set of behaviours in different networks. Identifying these subgroup activities will provide multiple channels for revenue enhancement activities.

A typical distribution of type of interactions among the subscriber groups is shown in Table 1. This indicates that the cross sell and acquisition is well suited problems to be tackled by MSNA.

How MSNA is different from Statistical Modelling

Table 2: Difference between profiling based on domain expertise, statistical modelling and MSNA

MSNA is based on the social relations than just individual behaviour as studied in statistical modelling or the selective profiling based on domain expertise. The following are the basic differences in the approaches:

How MSNA is different from SNA

Though MSNA has origins in SNA, the depth of coverage and amount of data crunching makes MSNA to be an independent method compared to SNA.

Table 3 compares the two methods.

Extracting the Influence

A profile of each user in the network called social proximity index can be used to drive lot of activities for campaigns. A social proximity index is a composite index which contains multiple measures combined in a weighted sum or some other proprietary form to maximize the information utility. 

Addition of the social network metrics in the behavioural modelling not only improves the accuracy of prediction, but also improves the homogeneity of the social network identities. This homogenisation of the subgroups will result in better information for predictive modelling than just behavioural data from aggregates. The most important metrics derived from the social groups, which can be used independently or in predictive modelling are:

Centrality: It has been observed that the subscribers who are in periphery of the network are least connected and are not useful in conducting any experiments.

Influence: It is often sufficient to condition the influencers, who in turn will effectively influence the connected subscribers in their immediate network.

Duality: Most subscribers have different behaviour in different networks as a group or as an individual. This helps to identify the cross sell/up sell opportunities.

MSNA in Campaign Analytics

A recent research has indicated that certain types of behaviour exist in a social network. For example, the survey results in Table 4* show that the ringtone recommendations are predominant with both friends and acquaintances. This is a good starting point to identify the potentials in the social subgroups for cross selling/up selling specific ring tones within the social networks.

In addition, this helps in identifying the psychographic profiles of both the individual and groups, which can be extracted from a combination of behavioural modelling and social network modelling.

An interesting input for campaign design can come from the evolution of the social network over a period of time. When the social metric for each subscriber is scored periodically and analysed, it throws light on the influencing behaviour of the few subscribers within the network causing the growth or shrinkage of network. Such analysis can be used for early intervention and targeted campaign for retention or X-sell or acquisition for revenue enhancement.

*Reference: Giuseppe Lugano and Pertti Saariluoma, 2007. “To Share or not to share: Supporting the user decision in Mobile Social Software applications,” Proceedings of the International User Modelling conference (UM 2007; Corfu, Greece, 25–29 July). 

MSNA Architecture

MSNA requires massive storage and processing power for a short period. It is ideally suited to be deployed in elastic cloud. However, due to privacy restrictions, the CDRs cannot be processed over cloud.

Hence it requires massive infrastructure investments to handle the BIG CDR data. This inhibited the use of MSNA in telecom for a long time, though CDRs were available for long. However, the introduction of distributed computing technologies using commodity hardware has ushered new era in MSNA.

At present, MSNA solution providers use a Hadoop based stack for MSNA. The architecture is shown above. The CDR data will be pre-processed by a Hadoop-Hive based pre-processing engine, which provides the multiple flavours of data like, network, individual and aggregate for modelling. The network models are then developed using MSNA engine. Subsequently the scoring and visualization of the derived networks and its properties will be done by the scoring and visualization components of the solution. The entire stack is configured for quick deployment.

MSNA Methodology

Unlike the behavioural modelling, which uses aggregated data or social network analysis, which uses small world/sample/survey data, Mobile social network analysis requires detailed to data to build and analyse the network models. The process starts with pre-processing the CDR data to the required format.

Once the network models are built, the different social metrics are attached to individual subscribers, which can be further analysed using visualization or other applications or can be used for enriching the predictive models based on behavioural data.

Conclusion

Mobile Social Network Analysis is a powerful tool that should be there in every Telco’s arsenal. It provides a different and enriched view of the customer base in addition to domain based and statistical modelling approaches. MSNA requires careful selection of the hardware and software to implement. The recent advances in Hadoop based solutions have sparked new interest in MSNA. This paper has highlighted how MSNA can be used for increasing the WAR effectiveness in telecom. Further it has provided the architecture and approach for MSNA with typical applications in campaigns.

About the Author

Dr. Jay B.Simha is Chief Technology Officer, ABIBA Systems, a telecom BI & Analytics company based out of Bangalore. He has about 15 years of experience in R&D, Business Intelligence and Analytics consulting. He has implemented large scale systems for telecom, BFSI and manufacturing industries in Business Intelligence and analytics. Dr. Simha holds a Doctoral degree in Data Mining and Decision Support and Post Doctoral from Louisiana State University, USA. He is active in research and has interests in business visualization, predictive analytics and decision support. He has so far published about 40 papers in international journals and conferences in the areas of business intelligence and analytics.