The three big telco AI use cases—and a lot more sub-use cases

By Sean Kinney, Editor in Chief June 16, 2024

Collected at: https://www.rcrwireless.com/20240616/fundamentals/the-three-big-telco-ai-use-cases-and-a-lot-more-sub-use-cases

Telco AI forecasted to become a $42 billion business by 2033

While artificial intelligence (AI) hype is running rampant, AI is also bringing real benefits to real businesses and the future seems relatively bright. Focusing in on telco AI use cases, Tantra Analyst Principal Prakash Sangam, speaking at the recent Telco AI Forum (available on demand here), broke things out into three broad categories of use cases: network planning and dimensioning, customer and network operations, and radio link management. 

Sangam also shared an important reminder that, in this particular case, what’s old is new again. “AI is not new to telecom as such,” he said. “We’ve been doing smart things in the network, on the devices, in the connectivity, for a very long time. More than a decade, I’d say.” He described self-organizing network (SON) technologies as something of a precursor to AI. As for what’s changed, “We are actually using [generative] AI and the newer models and newer concepts to improve the utility and performance even further.” 

Right-sized networks and AI-driven digital twins

For network planning and dimensioning, Sangam said the broad goal is to right-size networks so they’re not under- or over-provisioned. AI can be used for site and coverage planning, capacity planning, forecasting growth in traffic demand, and better understanding spectrum needs. A longer-term goal would be to use all the data informing these processes to stand up a digital twin that would allow for virtual “deployment” and “operation” based on changing real world conditions. 

“Basic objectives and benefits here of AI are right-sized builds,” Sangam said. “As we all know, networks are built for peak capacity and that is only for a certain duration…So if you over-build, there a lot of unnecessary sunk costs for the operator. If you under-build, that creates not a great user experience…You have to basically balance that build just to meet the demand—not too much, not too little—and that’s exactly where AI shines.” 

AI for customer-facing and network operations

This particular telco AI use cases, which encompasses a lot of more specific operational processes, speaks to the two major challenges (opportunities?) operators are currently facing. They need to make more money more quickly, and they need to automate the operation of the network in a way that saves them time and money while also opening up new capabilities that could, in turn, lead to making more money more quickly. It’s a bit circular. 

On the customer operations side, Sangam highlighted AI-enhanced customer service, offers and service plans, advertising and marketing, customer provisioning, chatbots and troubleshooting. On the network operations side, there are also chatbot and troubleshooting applications, along with network optimization, mobility management, network slicing, advanced SON and energy saving. 

Customer-facing AI chatbots are “where the major part of the AI is being used right now,” Sangam said. On the network side, operators are inputting troubleshooting guides and other technical materials into AI models that are accessed by engineers via  a chatbot interface. “If you’re a technician, you get a trouble ticket…you basically, through a chatbot interface, you ask the questions to the models…instead of you manually searching through all the manuals and everything. The most probably cause and the experience of handling such issues previously are already presented to you very quickly.” Sangam, citing a conversation with the CTO of Telus in Canada, said they’ve seen a 20% improvement in technician productivity using this approach. 

Improving network performance and user experience with telco AI

Sangam said using AI for radio link management—things like channel estimation, determining coding, modulation schemes, etc…—is a focus area for standards body 3GPP with study items in Rel. 18 set to become work items in Rel. 19. The high level is, “How can we select the best beam for the users you’re serving,” Sangam said. “AI has a huge role to play there.” 

Sangam summarized with a few primary considerations operators should keep in mind: 

• How to move from open loop to closed loop automation.
• Where in the network to host compute for AI workloads, which is very much related to architectural trends around virtualized and open radio access networks ((RANs). 
• How AI fits into the adoption of new service management and orchestration (SMO) technologies and, in the context of Open RAN, the relationship with the real-time RAN Intelligent Controller (RIC) and non-real time RIC.
• Data access, sovereignty and ownership.
• The logic around using third-party large language models (LLMs), and other types of multi-modal foundation models, as opposed to building telco-specific models either in-house or via consortia.
• And the workings of an AI continuum that spans devices, edge and centralized clouds and the network itself.