The Trifecta Effect for Telco Analytics –Anomaly Detection

Subex recently participated in the Monetising Big Data in Telecoms World Summit 2018, Singapore, where we demonstrated our expertise of 25+ years in the Telecom domain handling data at a massive scale. We did this by presenting on the topic: The Trifecta Effect for Telco Analytics –Anomaly Detection

To take a step back, Subex has partnered with 250+ telcos across 100 countries. We have been handling big data and have an understanding of the business of telcos working in different contexts, demographics and geographies, and at different stages of their growth. We have been leveraging analytics for 10+ years in the assurance portfolio, and have made a foray into analytics across all domains in the telecom sector. While starting on this journey, we did a survey across many operators and what came out was unexpected albeit not exactly surprising, based on our experience.

We saw that, while the telecom domain today is at the forefront of innovation, the industry can be considered as a relative laggard in adopting analytics, when compared with other industries. Around 60-70% of the executives still lack relevant data for decision making. To top it all, where analytics is being used, around 60% of the organizations are still not very confident about their analytics insights.

Based on our market research and after multiple interviews we believe that the reason for these above problems can be classified along the following categories:

  1. Poor ROI
  2. Multiple and Complex Dashboards
  3. Long Development Cycles
  4. Data quality issues
  5. Short Supply of Data Scientists
  6. Lack of Agility

To cater to these problems, we at Subex have designed a way to address these issues. Our solution, ROC Insights stands on three pillars which we call the Trifecta of Analytics: Agility, Cost & Consumption. Subex does this by

  • Delivering insights in less than 8 weeks
  • Following a pure OPEX model which takes care of the cost
  • Most importantly ROC Insights simplifies consumption of analytics insights tremendously through the use of storyboards

At the Monetising Big Data in Telecoms World Summit 2018, we explain how the Trifecta can be leveraged by using a simple example of Anomaly Analytics. For a telecom, with massive volumes of data, it is more difficult to detect anomalies in data than finding a needle in a haystack. In case of the latter we know that we are looking for a needle. In this case, we don’t even know what exactly one is looking for. However, it is extremely important for a telco to manage anomalies to mitigate risk and to prevent missing out on opportunities.

There is a very fine line between the definition of an anomaly and outlier. Rather the distinction is rather fuzzy as anomalies and outliers are intersecting but certainly not subsets. Take for instance, a queen bee in a bee hive. She is an outlier but not an anomaly. Anomalies usually remain undetected. They are unknown problems with unknown solutions. Our work detects, curates and qualifies these problems to move it from the unknown-unknown realm to the known-known realm by breaking it into parts, analyzing them using various ML/DL algorithms and as well doing causal analysis to find the root factors.

We talk of two examples where we work with Telcos to solve their anomaly problems using advanced analytics. In the first case, our solution of anomaly detection required modification as the client was based out of East Africa with pockets of high population density in vast open areas. We developed anomaly for cell sites using algorithms such as time series, manifold learning, LSTM etc. We did anomaly detection for different KPIs such as call duration, data and customer latching. This was further qualified along the dimensions of 2G/3G/4G, on-net/offnet and so on. Anomaly analysis generally suffers from the problem of ‘too many’ and difficulty in prioritization. Our solution gives revenue numbers to the anomaly and prioritizes them. Most importantly it looks also at the long term business impact of high value customers, churn etc. and also considers the factors whether load balancing is done by nearby cell sites and the customers were really impacted or not.

Our second use case was regarding our work for a client based out of N. America. They were having difficulty in detecting lost handset on time. This was causing them huge monetary loses. With the help of algorithms like probabilistic graphical model, Markov chain etc. we created algorithm to detect whether a handset was stolen or not. Our algorithm helped in improving the client’s solution 9 times in detecting the algorithm. As well in 80% of the cases, the instances were detected within 24 hours, thereby improving the customer’s bottom line.

All in all, the presentation provided for a good opportunity for attendees to understand how Subex is working with big data, and how ROC Insights can help telcos by pinpointing upon a very specific problem. At large, we enjoyed presenting at the event and meeting with telcos from across the APAC region. Hope to see you there next year!

Dr. Sumit Singh is Senior Data Scientist at Subex Digital LLP. He has a PhD in Decision Science and is a Machine Learning researcher, practitioner and educator with over 10 years’ experience in academia and the industry. At Subex, he is responsible for statistical modelling and algorithm design for business requirements. As well he is involved in doing research activities to implement new ideas into telecom domain. His areas of interests are – Reinforcement Learning, Optimization and Stochastic Modeling. He can be reached at

Related Post

Get Started with Subex