This decade started with the wave of big data. Big data technologies have made it possible to manage and process data-volumes and data-velocity of the order that was considered impossible earlier. Big data technologies have been used in several domains like social-media, retail, advertising and finance. Here we will discuss about its application in telecom segment.
At a broad level, three things are required to apply big data analytics to a system. First of all, one requires access to the data which needs to be analyzed. Second requirement is knowledge of the domain (e.g. telecom domain). Third requirement is knowledge of data sciences. In telecom, telecom service provider has access to most of the data involved. Here we will discuss various usecases of big data analytics that a telecom service provider can deploy to churn out the hidden information (or insight) within its network data and subscriber data.
A network infrastructure put in place to provide voice and/or data connectivity using wireless or wireline technologies is referred as telecom network. 2G/3G (wireless) based voice and data connectivity is an example of telecom service. DSL based data connectivity is another example of telecom service. In this article, wireless mobile network is taken as an example.
In a typical scenario, a subscriber initiates a service usecase using his terminal. The terminal connects with access network over radio signal. Access network connects with core network through backhaul connection. Core network carries out required connections with other subscriber, another network or a gateway, as required by the service usecase. While operations are on, operators of the telecom service provider can monitor performance of the network using Operations Support System. A subscriber can also place an new order to activate a new service or terminate an existing one by communicating with an operator (using voice-call, SMS, or email). Once an operator receives the new order, he/she initiates the respective workflow through Operations Support System to fulfill the order.
In these scenarios, broadly there are primarily three kinds of data which can be streamed into a big data analytics server to churn out useful trends and insights, hidden inside the data. These are as follows:
Like retail services and financial services, telecom services also have active involvement of its end-customers. The data generated in telecom services can provide useful insights about customer behavior, network performance and trends related with varous services. Traditionally, only offline business intelligence tools were used to churn out these insights. That also had a limited use because of high volume of data being continuously generated.
With emergence of technologies like Hadoop framework and NoSQL databases, it became far more affordable to run analytics on large set of data. These technologies facilitated running applications like recommendation engine to determine most suited bill plan for a subscriber using charging data records of the subscriber from past few months. However, a new revolution in telecom analytics has started with arrival of big data technologies which facilitate real-time analytics on these data streams. Some of these technologies are Apache Storm, Apache Kafka, Apache Samza, Spark Streaming and Amazon Kinesis. Real-time analytics facilitates churning out useful insights in real-time such that a proactive action can be taken to improve customer experience, which gives a boost to customer loyalty and avoids potential loss of business.
Some of the usecases from telecom networks are as follows:
A 3G mobile handset has two radio-access settings; high bandwidth and low bandwidth. High bandwidth consumes more power and low bandwidth consumes less power. When mobile handset is used for data service then it goes into high bandwidth setting. In traditional design, return of the mobile handset to low bandwidth setting is governed through expiry of timer. However, through run-time analytics of telecom service provider, it can push new carrier settings to the respective mobile handset, to switch it back to low bandwidth setting.
Figure 2 illustrates that in both cases transition from low bandwidth to high bandwidth happens when mobile handset is used for data services but in real-time analytics based design, transition back to low bandwidth happens much before expiry of the timer. This results in saving power in batteries of mobile handsets, everytime when data services are usual.
This energy saving in one mobile handset may look insignificant but saving this much energy for millions of subscribers and that too daily on on-going basis can make a significant difference.
In the network of a telecom service provider, at every moment some of the subscribers face one problem or another. Sometimes, voice quality is not good. Sometimes, there are call drops. Sometimes, data connectivity is very slow. While majority of the subscribers might be happy with the service but the service provider always carries a risk of losing this small fraction of subscribers.
In traditional approach, the service provider comes to know about these issues only after complaint is received from subscribers. Even if service provider uses traditional business intelligence tools, then also this information can be churn out only offline, for example, in a cycle of once in a month.
With big data technologies, telecom service provider can deploy a real-time analytics solution which can monitor quality-of-service in real-time. Based on signaling and content exchanged the analytics server can determine in real-time, what fraction of subscriber-base is facing service-related problem. Further it can also categorize the problems. For example, which fraction is facing signal drop and which fraction is facing data connectivity issues. The system can facilitate real-time investigation of root cause of each problem. For example, if majority of the problems are related with data-connectivity then using real-time analytics solution, the service provider can figure out how many issues are due to coverage problem, how many are due to congestion, how many issues are due to a misfit service plan and how many issues are due to subscribers handset.
Based on this insight, the service provider can take action for resolving coverage and congestion related problems, even before complaints start pouring in. Similarly, the service provider can reach out to the remaining problem-facing subscribers with a proposal to change their service plan or handset, such that their overall experience can improve.
It is commonly observed that when a subscriber goes on roaming from local network to a network in a different geography, then often there is a significant drop in his/her usage pattern. Very often, such traveling subscribers limit themselves to basic signaling, in order to avoid high-cost associated with international roaming. This results in significant loss of potential revenue of the telecom service provider.
Using a real-time analytics solution, a telecom service provider can predict whether a subscriber is expected to go silent when he travels to a new geography and for how long he is likely to stay in the new location (by analyzing historical data and recent events).
This can help anticipating whether a subscriber is in new location, for just a transit of few hours or he/she is expected to stay there for several weeks. For example, if he/she has been in London for more than 24 hours, then he/she is expected to stay there for two weeks. Accordingly, the service provider can device attractive roaming offer, for the traveling subscriber such that he/she finds a valuable proposition in using his/her subscription for his/her communication needs. Such attractive offers might be given to only those who are expected to go silent and have stayed in the new location for more than 24 hours.
This insight provides an opportunity for a win-win proposition between the service provider and the subscriber. This approach will help the service provider in minimizing its potential loss of revenue. On the other hand, subscriber will have a better experience because he/she will not have to explore alternatives if he/she gets a good proposition from his/her regular service provider. The service provider can also use this approach to change the subscribers behavior over a period of time such that he/she gets into habit of using services of his regular service provider during travel, even when an attractive offer is not made.
Self organizing/optimizing networks (SON) have gained popularity since LTE standardized its objectives. One of the objectives is mobility load balancing. This means that when there is congestion on one base-station and it starts under-performing, then the network should automatically reconfigure itself such that part of the load is diverted to neighboring base-stations and overall network performance remain healthy all the time.
In such systems, an operator configures idle/active handover parameters, cell-neighbor-pair parameters and user specific parameters in load balancing algorithms running on an analytics server. The analytics server receives status of base stations from network probes, which are spread in the network. The analytics server runs load balancing algorithms and determines optimal configuration for the base stations based on received status.
Whenever the analytics server proposes a change in configuration of any base station (after running its algorithms), the changed configuration is provisioned on affected base stations. This makes sure that the system automatically discovers under-performing base stations and its neighboring base stations are subsequently reconfigured, such that part of its load is diverted to them.
In this case, analytics server runs its algorithms on current data reported by network probes and compares it with historical data. By analyzing past outages and the network conditions which led to those outages, the analytics server predicts upcoming potential outages. This helps network operators to preempt the problem and take precautionary steps in advance, such that the outage can be avoided or its impact could be minimized.
Big data analytics has benefited several domains like retail industry, finance industry, etc. Telecom service industry is no exception. However, due to the nature of service that telecom industry provides, analytics solutions in this domain has some flavor of its own. Here, same subscriber uses the service several times in a single day itself. Moreover, a huge infrastructure of network-equipments is maintained by the service provider to provide un-interrupted and seamless service.
Due to these dynamics, analytics in telecom domain is not limited to just recommendation engines in decision support systems, rather there are several usecases where real-time analytics can be applied.
1. Hughes Systique , Real-time analytics with Apache Storm, http://www.hsc.com