From fraud prevention to RAN energy savings, Rakuten Mobile is using AI to move network data from insight to automated action
The telecom industry discourse on AI keeps returning to a familiar point: there is no AI without data. True enough, but not quite complete. The more useful formulation is that there is no valuable AI without the right data, tied to the right use case, producing a measurable business outcome.
Sachin Verma, chief data and AI officer at Rakuten Mobile, framed the company’s AI work not as a generic network modernization exercise, but as an ecosystem-level discipline. “This is all driven by what kind of a use case you want to solve with the data and the insight that we generate,” he said.
That shows up in places that do not look, at first glance, like network operations. Verma pointed to Rakuten Card, where network-level signals helped identify signatures associated with potentially fraudulent transactions earlier than the card team could see them. In a two-week pilot, he said, the card team was able to reduce riskier transactions by 33%.
He gave a second example around Rakuten Pay. Using aggregated, privacy-preserving mobile payment insights, Rakuten Mobile could identify locations where mobile payments were happening but Rakuten’s merchant footprint was weak. Passing that information to the payments team helped improve merchant onboarding in specific areas and increase business revenue.
The point is not that mobile network data magically becomes fintech strategy. The point is that a telco sitting inside a broader digital services company has a different surface area for AI. Data becomes more valuable when it can move from detection to decision to business action. This entire approach is also highly-instructive to service providers looking to reinvent themselves as platform companies built atop mobile networks.
But the same logic applies inside the network. Rakuten Mobile announced in February that its autonomous energy efficiency work, developed with Rakuten Symphony, had been validated by TM Forum at Level 4 for the RAN Energy Efficiency Optimization scenario in a live Open RAN environment. The company said the solution delivers around 20% RAN energy conservation through autonomous operations.
Verma described the architecture in practical terms. The RAN Intelligent Controller (RIC) hosts rApps that use machine learning to predict radio site behavior and take action in a closed loop. “In this case, a human doesn’t take any action,” he said. “A machine learning algorithm reads the data, takes the decision, and executes the decision.”
That is the difference between automation as a tool and autonomy as an operating condition. The system knows when the network should be on, when it should be off, for how long and under what conditions. Verma said the Level 4 use case has been implemented across Japan, reducing power bills by 17% to 20% and producing “close to a billion yen yearly saving.”
Rakuten has also completed nationwide deployment of RIC applications across its commercial mobile network in Japan and expanded third-party rApp integration with partners including AirHop Communications and Future Connections, positioning the RIC as a real-world platform for predictive maintenance, mobility enhancement, traffic optimization and AI-assisted decision-making.
The next challenge is end-to-end autonomy. Verma said Rakuten’s “first line of defense is all done by agents,” but the company is still working toward a broader fabric that correlates alarms and events from radio sites through IP transport, core, cloud and platform layers. “The first portion is detection. The second portion is analysis and the third portion is remedial action,” he said. “We are good with the detection piece. We are good with the root cause analysis. I think the remedial action is something that we still work on.”
In the broader move from automation to autonomy, Level 4 in one domain does not mean Level 4 everywhere. The agentic network will be built use case by use case, closed loop by closed loop, with measurable value as the forcing function.
And, as with every operator pursuing AI-native operations, the technology problem is also a people problem. “This is one of the biggest hurdles for the adoption of AI,” Verma said. “The technology is growing at a pace, but the diffusion of technology in an enterprise is pretty low.”
His prescription is top-down change management and a clear message to employees: AI is not simply taking work away; it is changing the work. “It is not about AI taking the job,” he said. “It’s all about AI enabling you and making you more efficient.”
The Rakuten Mobile lesson is that agentic AI becomes credible when it stops being abstract. Reduced fraud exposure, lower power bills, faster mean time to repair and a more productive workforce can all be measured. The network does not become agentic because an operator deploys agents. It becomes agentic when data, AI and operating processes are disciplined enough to turn intent into action.
