Rakuten Doubles down on AI
Japanese operator Rakuten is accelerating its “AI-first” strategy, setting an ambitious goal to achieve Level 5 of the TM Forum’s “Autonomous Network (AN) Maturity Model” within 18 to 24 months.
Sachin Verma, Chief AI and Data Officer at Rakuten, stated: “We have self-assessed that we have reached level four in certain scenarios, but the maturity curve is exponential—from 0 to 70 quickly, yet 90 to 95 is extremely challenging. Therefore, 18 to 24 months is a realistic target.”
The timeline aligns with industry trends: a recent Bain-TM Forum survey indicates that 20% of operators have achieved Level 4 or 5 in certain areas, while 35% expect to reach these levels within two years.
Currently, no operator worldwide has officially achieved Level 5. The TM Forum’s AN assessment categorizes network autonomy into six levels: Level 0 (fully manual) to Level 5 (fully autonomous).
Sachin Verma, Chief AI and Data Officer at Rakuten, which serves around 9.08 million mobile and MVNO users, described himself as an early AI advocate and adheres to an internal quantifiable “Triple 20” AI principle: any AI initiative must deliver a 20% efficiency improvement in marketing, operations, and customer service.
Its AI strategy focuses on four key areas: customer experience, network, operations, and sustainability
In terms of customer experience, AI first identifies the scope of faults, then automatically calculates compensation plans, and schedules neighboring antenna tilt angles to ensure coverage. “Fully self-developed, AI is integrated at every step.”
On the network side, AI focuses on three key KPIs: quality, coverage, and congestion. Among 400,000 indoor and outdoor cells, AI proactively identifies high-traffic events and dynamically expands capacity.
In terms of energy efficiency, Rakuten rejects the practice of “shutting down during idle periods” to avoid compromising user experience, opting instead to reduce power consumption in low-traffic areas. This model has been implemented across more than half of the network, achieving energy savings of 17% to 22%.
At the operational level, AI is utilized for anomaly detection, root cause identification, automated resolution, and closed-loop management. The company is piloting several AI agents with an accuracy rate of 85%, currently adopting a “human-machine hybrid” model.
Verma cautioned: “AI agents must carefully calculate the cost: while the unit price of tokens has dropped, their usage has increased, potentially saving $10 but costing an extra $40.”
Rakuten has initiated the development of its proprietary “Small Language Model (SLM)”, which requires significantly less computational power and infrastructure compared to large models. It offers faster deployment, easier optimization, and the ability to retain data locally, eliminating the need for GPU during inference, resulting in superior cost efficiency.
Earlier this year, Rakuten launched the “Rakuten AI 2.0 mini”—a dense model with 1.5 billion parameters and hybrid training in Japanese and English, specifically designed for low-cost edge device scenarios.