Demand forecasting and anomaly detection are AI capabilities that Google sees as accelerating numerous additional use cases.
A very material amount of all internet traffic, around 60% to 70%, runs through the Google Global Network. In reaching that scale, the company has learned a lot about using AI tools to manage the network with a high degree of automation, according to Naresh Rao, Google Cloud’s head of telco analytics. And now, he said, CSPs are benefitting from those same as they continue on their own network automation journeys.
Rao said use cases like demand forecasting, anomaly detection, root cause analysis and field operations management “have been [of] paramount importance to Google. For CSPs who continue to invest in network transformation against stagnant or declining revenues, “The most important aspect…is how to leverage AI…to optimize their entire network operations…and also improve customer experience.”
Rao spotlighted Google’s AutoML service, which is a set of machine learning solutions meant to enable developers to train models tailored for specific business needs. He said AutoML can deliver a 25% improvement in demand forecasting which opens up a variety of use cases, including fraud detection, network planning and predictive maintenance among others.
Expanding on predictive maintenance, he said proactively addressing potential network failures minimizes downtime, streamlines operations and also benefits end users. Rao gave the example of European CSP who began using Google Cloud services to ingest RAN telemetry, built a proactive model and generate a simple answer—can an issue be fixed remotely or does it require a truck roll. “That solved a lot of issues,” he said.
Anomaly detection is another area that “is more than a use case,” Rao said. “It is just a technical capability” that can serve as the foundation for numerous use cases. As Google’s own experience has validated, “Anomaly detection can run and scale…This is directly available to our telcos when they run their workloads on Google Cloud…It will help them to build more and more robust use cases.”
The big picture, Rao said, is realizing intent-driven networking. “A user will actually define an intent—‘This is what I would like to achieve on a particular network.’” That requires the ability to ingest a range of network telemetry from multi-vendor networks across multiple domains. That feeds into a common data model that feeds data pipelines. And, “It should be API-driven,” he said. “That’s what is very important.”
Rao reiterated the importance of inventory management in “achieving this North Star of autonomous operations…understanding all the network functions, network applications, the services and everything, so you continue to drive this intent until you achieve the desired intent.”
For additional reading on telco AI solutions read this interview with Fujitsu and this interview with Blue Planet.
