21 JUNE 2018

From controlling costs and assuring services to management networks – machine learning is critical to the future of telecom

While the term ‘machine learning’ was coined by Arthur Samuel in 1959, it is only now nearly 60 years later that we are seeing it evolve from academia to real-world applications. In fact, machine learning has already entered the homes of many of us. Netflix recently commented that its machine learning algorithm, which recommends personalized TV shows and movies to subscribers, saved the company $1 billion in customer retention costs by being able to quickly and accurately provide recommendations that keep their customers tuned in and coming back for more.

In 2018, the growth rate for Machine Learning and Artificial Intelligence is expected to be 70 percent, according to Gartner. And telecom carriers are prime candidates to reap the rewards, as they already have the fundamental 3Vs of big data: velocity, variety, and volume, and have become highly adept at storing, analyzing and processing these bits and bytes.  Customer data, location, network performance data, network traffic data, billing data and social media data are just a few sources. Now with machine learning, valuable insights within these data lakes can now be more fully understood and leveraged across the entire carrier organization. Here are some ways that communication service providers (CSPs) are utilizing machine learning today.

Customer Churn Predictions
Using Machine Learning capabilities, CSPs can now zero into a customer segment to understand if there are underlying network conditions that are impacting customer tenure and use this information to help drive future business value. TEOCO recently supported a CSP customer to take a deeper dive into their prepaid subscribers, whose average tenure was less than six months. They wanted to understand why their churn rates were so high.  We were able to go in and analyze data from a variety of sources across the entire business. This information included 2G, 3G & LTE xDRs, switch and cell reference information, rate plan data, plus subscriber reference and account activity, DPI session data and billed events. TEOCO was able to bring all this data together and analyze it, determining what happened moments before a subscriber churned, identifying critical location-specific issues, such as in-market roaming, dropped calls and 4G capacity limitations. These were all impacting the subscribers’ quality of experience. But to fix the problems, the carrier first needed to detect and understand what and where they were coming from.

Our use of machine learning helped quickly determine that the root cause was due to handset configuration issues failing to select the home carrier, along with LTE cell capacity limitations. Once rectified, service quality improved and the CSP reported that 90% of pre-paid subscribers increased their average tenure from 6 months, to over 13 months, representing a significant cost savings to the operator.

Revenue Assurance: Keeping Costs Under Control
While most service providers have experienced revenue assurance teams in place who are focused on managing network costs, these skills can be hard to find and are often in short supply because they take years – even decades – to develop. Machine learning brings a whole new set of tools to the table.  By creating algorithms that analyze and learn from past actions of experienced teams of financial auditors and revenue assurance specialists, opportunities for cost reductions can be found more quickly and efficiently as their best-practices can be replicated across other projects.

For example, TEOCO recently performed a financial audit that looks at how a carrier’s circuit installation charges were assessed. TEOCO was able to train its software using historical data from other successful facility audits, and the results have been impressive, with over 500 additional billing disputes identified by machine learning to-date, worth over ~$250,000 in cost recoveries. The software was able to identify circuits where the carrier was assessed an install charge, on orders that weren’t eligible.  By identifying these billing errors quickly, TEOCO was able to prevent future losses from piling up.

Service Assurance for Smart Cities
Advanced techniques in analytics and machine learning can also be used to help CSPs optimize new IoT offerings, including Smart City applications. Whether it is an airport, hospital, connected car, telecom infrastructure or entire Smart City, machine learning is critical to providing advanced performance management, automation and orchestration functionalities such as real time performance management, including KPI monitoring and real time threshold crossing alarms, event and trouble ticket management, trend analysis and forecasting and real-time outlier detection.  In the similar vein as to how CSPs leverage, correlate and analyze the massive amounts of data that traverses their digital networks, as our word becomes more connected, Smart Cities can integrate and correlate data from transportation, utilities, city services, weather tracking systems and social media, for example, to provide fast, coordinated responses to emergencies, natural disasters and terrorist attacks effectively and more quickly.

Drone Traffic Management
Machine learning capabilities are critical as more drones take to the sky, become truly autonomous and operate beyond visual line of sight. Managing drone traffic requires superior service quality and seamless coordination between regulatory authorities, radio space, airspace and communications service providers. Machine learning capabilities are critical to assuring drone connectivity for the duration of a mission as they cross Wi-Fi, 4G and 5G networks, to also understand performance at a service level in real time, and proactively detect poor performing devices, services and users and isolate the root cause of an issue quickly. At its full potential, machine learning can be used to optimize drone routes based on real-time weather update feeds, while protecting the safety of manned air traffic, humans and property.

All Roads to Prescriptive Analytics
Because most service provider data is now being housed in data lakes, the ability to extract greater value from this data is almost limitless.  Professor Bimal Kumar Roy from The Applied Statistics Unit at the Indian Statistical Institute sees machine learning as a crucial step towards enabling CSPs to predict future business value, increase profitability, grow revenue and assure network performance via prescriptive analytics.

“Machine learning is accelerating the ability for CSPs to predict future business value more than ever before. As CSPs look to launch service critical applications such as remote surgery, CSPs need to the ability to pre-empt any service impacting event before it occurs, and we see machine learning with prescriptive analytics being critical to achieving this. We’re just seeing the tip of the iceberg on how we can utilize machine learning in the telecoms space,” said Professor Bimal Kumar Roy.

Professor Bimal Kumar Roy is providing his expertise and mentoring TEOCO’s machine learning team, which is focused on developing new use cases and capabilities to support the world’s largest CSPs to optimize their business and manage their networks, services and applications in the digital era.