Telecommunication systems are coursing with time-series data. For example, a mobile operator with millions of subscribers will generate tens of millions of call records daily and billions of rows of call data over the course of a year. In turn, each of these calls is routed through a myriad of network nodes, switches, routers, servers, and other equipment, each of which generates huge volumes of telemetry data that are organized by time. Telcos now carry about 200 times more daily Internet data traffic than voice traffic (calls), and the total traffic that telcos are dealing with has increased over tenfold in the last five years alone. You can truly say that telecom companies have a big data problem — and opportunity!
The time-series data that telcos have at their disposal is a veritable goldmine of riches once Machine Learning is applied. Hidden in this treasure trove are insights that can help telcos:
- enhance customer experience,
- improve their network quality,
- and capture and retain more subscribers.
Testing customer loyalty with experience issues
As a telco operator, your subscribers expect your incredibly complex and expensive network to just work. Of course, most of the time, it works great — but subtle, hard-to-detect anomalies can impact your service quality and negatively affect the customer experience of your subscribers. These issues test your customers’ patience and could potentially lead to subscriber churn; as a result, loyalty refunds, credits, and incentives to keep them happy eat into your profit margins.
Time-series network analytics provides the ability for network operations and engineering teams to predict trends and detect anomalies in the network performance and proactively improve the service quality of their networks. Using predictive analytics on the network telemetry — including network performance, Internet usage, clickstreams, and traffic-flow data — enables operators to identify underperforming components of the network and improve the quality of experience for subscribers before they may even realize there is an issue.
Based on: Predictive Analytics for Time Series with InstantML, by L. Miller and Tangent Works (2021).