AI In The Network – An MWC Debate

*Guest Blog by Keith Dyer, Editor of The Mobile Network

At MWC 2025, I was honoured to host a panel on the Mavenir booth, looking at how operators are introducing AI technology across their networks and in their operations. It was a rewarding panel to moderate because it did one thing that can be unusual in industry panel debates.

Often panels like these start out with the technology, then approach the customer benefits as a final outcome, almost like an afterthought.

With AI this would have been a very easy thing to do – listing out all the wonderful things AI can achieve, diving deep into the capabilities that new models, Gen AI and Agentic AI (new buzzword for 2025) can bring. And of course with AI and Machine Learning there are advances all the time data science – there’s much to get excited about.

But this panel went the other way. It started with the customer, and then moved to very practical examples of how AI is being used in mobile network operations, and how Mavenir is bringing AI capabilities across the network. Finally, it looked at how a data-first approach can underpin those customer-centric applications.

Watch the Panel Discussion:

In his very first answer Tore Kristoffersen, from Norwegian operator ICE, said that the use of AI is driven by how it would help the customer. And he had a very real example ready – SpamShield, an AI-powered messaging fraud detection and prevention solution that has already blocked eight million messages in amongst the billions that ICE handles. That adds to the trust that ICE’s customers have for the operator. “We need our customers to trust our services, that’s what we live from.”

Practical Applications

A valuable aspect of the panel was how it concentrated on practical applications of AI and ML technology within the telco business and network operations.

Dimitris Mavrakis of ABI Research said that what operators want to do is simplify operations to create a better user experience. That means there are many models being introduced in the network, with predictive AI, Gen AI and LLMs being developed to help network operations become more efficient.

Of course, that means understanding which AI techniques can help in which circumstances, and being able to apply that to telco-specific data.

Mavenir’s Brandon Larson echoed Kristoffersen’s customer-first approach, taking a practical approach to the use of AI capabilities. Mavenir’s implementation of AI is not technology-first. Instead, it looks at what it can help its customers with and then works back to how AI can enable that.

Larson explained how a technique such as DQN Reinforcement learning is used by Google DeepMind and by Siemens to optimise cooling in data centres and gas turbines. Networks too are in essence complex optimisation problems. By using DQN reinforcement learning in an operator’s RAN network, and training it to tune mobility parameters, Mavenir found the AI outperformed the humans by 40% in a week.

Nor do telcos necessarily need to invest hugely in new hardware platforms to support these AI use cases. Larson said needing to put GPUs in the network to support the RAN tuning application would have been unsustainable. Instead Mavenir had to put trained AI models on available compute and cloud real estate, the existing management cluster and CPUs, training the models offline and introducing them in a commercially viable way.

“When we talk about AI it’s not all LLMs,” Larson said. SpamShield uses a classification model, for example.

Putting data first

The introduction of AI into network processes raises the issue of how telcos can combine telco-specific domain knowledge with AI and data science.

Kristoffersen said for ICE its approach starts with an architectural change – putting data first. “We try to be modern, attacking the market in the right way. So we moved to a cloud-native solution, a data-driven approach so that all the functions in the 5G core are data driven, and that’s when we can really make the magic happen between the data scientists and telco.”

For Mavrakis, there’s no doubt that structuring and accessing data will be key for future success.

“There is a lot of opportunity. The key differentiator telcos have is that they only use a fraction of the data they have access to. Every sector is trying to use data, and telcos have a lot of data which can be used in a compliant manner to train models. I think that can be a catalyst to push them forward.”

Larson agreed that a data-first approach would lead to use cases where AI can bring the most value.

There are a lot of problems in telco that can be solved with AI. There’s so much data to be crunched in data, that means there’s so much value to be extracted. The operators that lean forward and embrace it are going to have the competitive edge, no doubt about that.”

Conclusions

For me, what the panel showed is that those operators who think through their customer problems and then apply AI to the data they have can be successful. Of course, that approach raises challenges. Operators need to harmonise and structure their data, and they need to be able to apply AI technologies in a sustainable manner. That’s where Larson’s points about Mavenir’s practical, cross-domain approach made sense to me. There’s no doubt we’re truly entering the AI era in telecoms, and it is the companies who understand what they want to achieve with AI that will succeed in extracting the most value from the technology. And that takes us back to the start, and to the customer.