By Sean Kinney, Editor in Chief November 27, 2024
Collected at: https://www.rcrwireless.com/20241127/fundamentals/the-big-telco-ai-challenge
Google Cloud sees telco AI “absorption rate” as an “incredible phenomenon”
Communications service providers (CSPs) are all-in on AI; makes sense given macro issues around a lack of effective network monetization despite a massive capital outlay which is putting pressure on automation as the primary path to opex reduction. The recommendations from the vendor side, as CSPs embark on what’s likely a decade-plus long AI-enabled network and operational transformation, is to focus on use cases which themselves hinge on data, and make incremental technology decisions while bearing in mind the holistic goals. And that doing those things successfully will require a larger ecosystem than operators are accustomed to cultivating and managing.
In a panel discussion at the recent Telco AI Forum 2.0, available on demand here, Google Cloud’s Jen Hawes-Hewitt, head of strategic programs and solutions for the Global Telco Industry business, said her focus is building out a partner ecosystem and “getting sleeves rolled up, implementing some of these AI use cases.”
Discussing adoption of AI by the telecoms industry, she called it an “incredible phenomenon…AI has entered the boardrooms…faster than any other kind of technology shift we might’ve seen before that.” Hawes-Hewitt drew the distinction between CSPs experimenting with AI as opposed to moving it into production; Google Cloud is seeing an emphasis on the latter—”real, concrete, live, in-production use cases across whole swaths of their business process, and the measurement of the value against kind of key performance indicators.” She said the use of telco AI solutions is “advanced more so than the kind of general enterprise landscape…I think we should be excited by that.”
In terms of specific use cases, Hawes-Hewitt called out a range, including network planning, root cause analysis and multi-modal field technician assistance. A good deal of how Google Cloud approaches telco AI, she said, is based on the company’s own learnings in managing its massive global network. “That has really created those principles, autonomous principles, from the beginning for us.”
Looking at work it’s done with Telus’s field technician organization, Hawes-Hewitt said that allowing for voice and additional modalities to help field techs “quickly refer to a manual…[and] interact with an assistant.” The ability to use natural language and visuals is important, she said, for techs who may not be in a position to type something on a tablet. “This is real adoption.”
Before diving into the AI of it all, Nokia’s Jitin Bhandari, chief technology officer for Cloud and Network Services, took stock of the current state of affairs, specifically impending deployments of 5G Standalone (SA), then 5G-Advanced. “We are still in the early days of 5G,” he said, predicting a “huge amount of rollouts” of 5G SA in 2025. The implementation of cloud-native networks and management practices, along with enhanced cross-domain observability, sets the stage for “the notion of a construct of automation and autonomous decision making.”
“If you want to get to autonomous decision making, AI becomes a very effective tool,” Bhandari said. He also pointed out that CSPs are effectively using machine learning, or classic AI, quite extensively today; the use of gen AI is also quickly ramping. With a wealth of real-time, near-real time and non-real time data, both structured and unstructured, CSPs have the baseline they need to push forward to conversational network operations and agentic AI systems. All of that is going to happen, he said, but the technology stack “has to be born in the cloud.” And, Bhandari added, “You’ve got to have a very holistic approach” to data. Getting AI right “requires a lot of data science.”
While “It’s like 1,000 flowers blooming,” telco AI opportunities bring challenges
Back to Hawes-Hewitt’s observation that AI is drawing fast, broad interest from operator organizations—this also means there’s a challenge around where to get started. “We have this kind of explosion of ideas, but the next question is kind of how do you move into production?” she said. This requires a systematic approach to experimenting with different AI-enabled use cases, cherry picking the experiments that deliver value, then moving into production, all with strong, consistent governance. “Picking the winners…is a really challenging piece at the moment, and how do we measure return on investment for these use cases?” she said. “It’s like 1,000 flowers blooming.”
Bhandari delineated three major challenges that each come with their own set of sub-challenges. First, and aligned with what Hawes-Hewitt said, is identifying use cases and mapping them to ROI and business value; this is something that can vary quite dramatically from operator to operator depending on their scale, he said. Next is technology selection—primary considerations include on-prem or public cloud and open or closed foundation models. And finally, data. He described three layers of CSP data: data in networks, data in operations and data in the IT estate. “The fabrication of data in all these three layers is very, very different,” he said. “There is a lot of learning yet to be done in this industry…This is one of the very unique verticals which has got a large, varied set of data from real-time to non-real time, both structured and unstructured.”
Leave a Reply