Customer Experience Management Using Big Data Insights By Telecom Operators

It’s a challenging and exciting time for communication service providers. On one hand there is exponential increase in mobile data traffic, driven by newer trends/newer applications

  • Location aware mobile applications
  • Growing use of social media through smart phones
  • Ever growing demand for streaming video/rich consumer media content
  • Streaming traffic from sensors, IoT, M2M surge etc.

At the same time, the industry is witnessing other trends that complicate matters: ever growing customer expectations, saturated market (stagnant ARPUs) and customer acquisition becoming more and more costly.

The key to transcending these challenges is a strict focus on customer experience, which will help retain customers, minimize churn and build on revenue streams from new trends like IoT, M2M.

Leveraging big data models for customer care

One of the major applications for big data solutions is customer care. Historically, service providers were doing churn analysis only after the customer left the service. But now by employing big data based solutions service providers can proactively and positively influence the customer experience to pre-empt and minimize the possibilities of switch.

Today, Hadoop-based big data analytic tools have the ability to deal with a high volume of data, varied in nature (structured and unstructured) and received at a high velocity. These tools are able to store, process, slice and dice data in the context of specific subscriber and/or specific network element – all in near real time. Such capabilities are likely to go a long way to influence customer satisfaction and improve customer loyalty.

Let’s take a use case: A senior corporate executive subscribed to a premium 4G-service makes a service call to log a problem related to his mobile internet access. The initial problem statement from the executive to helpdesk is “intermittently poor quality while playing live channel from YouTube.”

If the operator hasn’t invested in big data technologies, the scenario would likely take the following course:

The customer representative logs the problem and it goes through a typical investigation cycle with no deterministic timeframe commitment to customer for issue resolution. When analyzed by the operations team on several occasions, they are unable to pinpoint the root cause and are likely to treat the issue as a transient issue. Often at this point, the customer continues experiencing the same issue and eventually leaves for another service provider.

But if the operator has invested in big data technologies, the scenario has the opportunity to follow a different course, with much improved results. In this case, the call-center representative would assign the issue to network operations team, promising a call back to subscriber on same business day with investigation results.

Prior to taking a deep dive into this scenario, let’s quickly look two important aspects:

  • Network Deployment: The operator network is comprised of four parts and the problem could occur at any of them:
    • RAN (Access Part)
    • EPC (Core Network)
    • IP Backhaul (connecting RAN with Core)
    • Service Network (Internet or IMS etc.)
  • Big data tools and capabilities: The network service engineer employs the Hadoop 2.0-capable tool in batch mode to investigate the “YouTube traffic choke problem.”  The figure below depicts a quick view of tools, capabilities and workflow. It is assumed that network KPIs, customer CDRs, ISP Router data and network probe data is stored in Hadoop Distributed File System (HDFS) and Network Deployment map is stored in a spatial database (e.g Open GEO)

With this background, let’s take a deep dive into the second scenario - the service provider has invested in big data based technologies. The network engineer, assisted by batch mode map reduce scripts, could execute the following sequence:

    1. KPI per Network Node: Script starts with analysis of KPIs from various network nodes to isolate packet drops/traffic bottlenecks seen at any of the nodes/interconnects in the network. At a high level all statistics were found to be clean.
    2. CDR Analysis: Next, script checks subscriber CDRs at PGW (Packet Gateway for analyzing CDRs of different classes of service like web browsing, real time/video, FTP etc.) of all active sessions at that point in time. No explicit problem was observed in CDRs for non-real time class of traffic, such as browsing or FTP sessions.
    3. Filtering for Real time Services: Zooming into real-time services (video streaming), the batch process filters all CDRs of video sessions. In some of the CDRs the traffic is not building up as per streaming video session.
  1. Overlay of Deployment View: To further isolate the issue, the service provider personnel overlays the deployment view (location of network nodes plotted in geographic map) on the filtered CDR stream.
  2. Merge Network Probe data: Next he pulls in network probe data. There is red dot visible w.r.t to inward traffic and it seems there is drop in volume at the point-of-connect of EPC network with the internet.
  3. Merge with Packet Flow Stats from ISP Routers Further analysis of the CDRs and network probe data, along with the packet flow statistics from routers of internet service providers (data sourced from vendor service portals), show that inward traffic is facing many packet drops in the case of one specific ISP.
  4. Filtering IPS-Wise: It is clearly visible that a router just prior to the point of entry for one ISP (ISP1) is showing several packet drops, resulting in the issues faced by all customers routed through this ISP (ISP1).
  5. Now as an immediate solution, he puts this customer into separate category where different ISP (ISP2) is employed, thus giving first level resolution. At the same time, he raises the ticket for vendor management team to work with this specific vendor to get the issue resolved.
  6. To close the loop, the customer representative contacts the customer, informing him that issue has been identified, initial resolution is already in place and the problem will be permanently addressed in next couple of days.

While the actual implementation of above use case may require ironing out further low-level details, the emergence of such use cases is becoming vital with the maturity of big data concepts and associated tools in order to meet customer expectation and minimize churn.

 

Leave a Reply

Your email address will not be published. Required fields are marked *