Helping customers before they call

Share this post:

by Francesco Calabrese, manager of Smarter Urban Dynamics, IBM Research-Ireland 
As our smartphones get smarter, we’re using more Over-The-Top applications like WhatsApp, Viber or Skype, rather than our telco provider’s voice, and text. This downward trend means shrinking income for the telco, even though it’s estimated that data usage will grow beyond 20 exabytes per month in the coming years. When my team of social, mobile and decision theory researchers at IBM’s lab in Dublin noticed this OTT trend, we wanted to know, in broader terms: could we measure and predict the quality of a customer’s experience on a telco network in real time?

Telcos need to deepen their customer knowledge through smarter capabilities such as predictive analytics.
– IBM Institute for Business Value (IBV) study, Restoring Connections
And we may have found the answer in predictive analytics. Our recent ArXivpaper, Towards Real-time Customer Experience Prediction for Telecommunication Operators, tells an interesting story of telcos’ inability to anticipate customer problems. If telcos could predict issues in real time, they could then serve their customers more efficiently by anticipating problems and implementing fixes before a customer calls with an issue. The good news is that our study found that telcos have the data necessary to help – and keep – their customers. 
So, we devised a way for a telco’s system to understand a bad customer experience so it could examine, and anticipate customer usage. This, in turn, would give the telco a chance to intervene in real time, in what the system considered a potentially bad experience in an automated way, such as a text to customers, or to send operators out to fix a technical issue. For telcos, this predictive analysis could not only mean improved customer satisfaction and reduced help calls, but less customer churn, and a better Net Promoter Score – a telco’s customer experience grade.
Piloting through a forest of data
We partnered with a major South African telco to test our ideas on some real, but historical and anonymized, data. They provided a data set of app usage, which we plugged into a big data architecture of network analysis, and machine learning software. We used the IBM Now Factory network analytics “feed probe” to collect the data and push it into Big Insights machine learning algorithms. This allowed us to generate models of contextual relationships within the data. And then we used InfoSphere Streams analytics platform to score such models on millions of events from the network and make predictions of customer experience. During the pilot, our application processed billions of records that included mobile app usage and performance, and network location. It then evaluated and visualized how these factors affected churn rate, customer care and Net Promoter Score. 
For example, we uncovered an issue with a mobile application that was linked to a high percentage of retransmitted data, which resulted in many customer care calls during the monitored days, as well as for two subsequent weeks. If the telco had deployed our predictive solution in real time, it would have been able to identify the issue well in advance, and proactively communicate it to the customers, thus reducing the number of customer care calls. 
Ernesto Diaz-Aviles, machine learning and data scientist, IBM Research-Ireland
The system discovered the issue by training what we call a Restrictive Random Forest model. It observed customer transactions in the telco’s data feed. It then generated an ensemble of decision trees that labeled each daily experience as either positive or negative, based on if a help call was made. This approach could allow the system to help prepare call center agents to manage a problem before a call, better manage a call that does come through, or tell technicians of a (potential) mechanical failure – before anyone experiences an outage.
My team, and our system, will continue to study the individual aspects of user behavior and the reasons for their calls. Our goal is to build personalized models for segments of users, an approach we expect to more-accurately predict the user’s mobile experience. So, maybe in the future, customers will make a call on their telco’s network for something other than reporting a complaint. 
More stories

A new supercomputing-powered weather model may ready us for Exascale

In the U.S. alone, extreme weather caused some 297 deaths and $53.5 billion in economic damage in 2016. Globally, natural disasters caused $175 billion in damage. It’s essential for governments, business and people to receive advance warning of wild weather in order to minimize its impact, yet today the information we get is limited. Current […]

Continue reading

DREAM Challenge results: Can machine learning help improve accuracy in breast cancer screening?

        Breast Cancer is the most common cancer in women. It is estimated that one out of eight women will be diagnosed with breast cancer in their lifetime. The good news is that 99 percent of women whose breast cancer was detected early (stage 1 or 0) survive beyond five years after […]

Continue reading

Computational Neuroscience

New Issue of the IBM Journal of Research and Development   Understanding the brain’s dynamics is of central importance to neuroscience. Our ability to observe, model, and infer from neuroscientific data the principles and mechanisms of brain dynamics determines our ability to understand the brain’s unusual cognitive and behavioral capabilities. Our guest editors, James Kozloski, […]

Continue reading