China Mobile’s new GenAI solution leads to $7 million in opex savings

Who: China Mobile and Nokia

What: Developed a GenAI solution for operations and maintenance

How: Leveraged TM Forum’s work on AIOps and autonomous networks frameworks to integrate an open-source large language model and local knowledge centers to assist engineers, improve performance and reduce opex

Results:

• 80% reduction in knowledge acquisition time
• 72% increase in data analysis efficiency
• Annual savings of $7 million in network operations costs

In its Guangdong, Beijing and Fujian provinces, where China Mobile has a total of 190 million subscribers, the company has been working with Nokia since 2022 to develop a GenAI solution to help with work order optimization, field testing and performance optimization of 4G and 5G networks. As a result China Mobile is saving about $7 million a year and getting closer to its goal of achieving a highly autonomous network.

Based on TM Forum’s definition, highly autonomous networks – Level 4 and Level 5 – mark the transition between traditional automation of human-defined process behavior and autonomous behavior, where systems make decisions independent of humans (see graphic below). In their quest to achieve these levels of autonomy, communications service providers (CSPs) worldwide have faced a challenge when it comes to implementing intent-driven operations, but China Mobile is demonstrating that the rapid development of GenAI could be a game changer.

The systems in autonomous networks are governed by intent, which sets expectations as requirements, goals and constraints. These are abstracted from the technical inner workings of the network. Put more simply, intents are the “what” not the “how” – meaning, you tell the system what goal or outcome is required without having to tell it how to achieve it. This decoupling increases agility, allowing autonomous systems to fulfill the intent using data, machine learning and AI.

By leveraging GenAI and large language models (LLMs), China Mobile and Nokia are achieving an accurate understanding of users’ intentions, which they then translate into intelligent orchestration of network operations. “The solution’s architecture is open, and its capabilities are decoupled, allowing flexible integration with different fundamental models to support numerous operations scenarios,” says Meng Liang, Chief System Architect of the Network Operations Center at China Mobile Guangdong. “The solution has demonstrated remarkable results in daily operation, effectively reducing opex.”

Addressing the challenges

China Mobile was looking to address three main challenges with the GenAI solution:

• Cut the time it takes to query knowledge bases when trying to resolve faults – a large amount of knowledge about telecoms network operations exists in various manuals, case libraries and reports, making it difficult to run queries. Indeed, it was taking China Mobile’s engineers an average of five to ten minutes to find information, with some complex cases taking more than an hour.
• Reduce the cost of daily data analysis – in operations and maintenance (O&M) there are many temporary and personalized data analysis requirements, which are labor-intensive and must be completed by highly paid experts. In the Guangdong province alone, there are more than 3,000 tables in China Mobile’s big data platform. With more than 400 engineers conducting data analysis tasks every day, the cost of data analysis was about $6.8 million a year.
• Improve DevOps development – managing diverse and dynamic business requirements in network operations demands significant time from software engineers, often spanning several days or even weeks. This results in low efficiency and high costs, with around 200 software engineers working in the three China Mobile provinces.

The GenAI solution developed by China Mobile and Nokia addresses these issues by providing a natural language capability for intelligent Q&A, automation of data analysis and DevOps code generation. To develop the solution, the team used TM Forum’s AI in Operations Framework, Autonomous Networks Framework and Business Process Framework (eTOM), along with an open and decoupled LLM framework based on LangChain, which is an open-source framework that enables the integration of LLMs with external components and data sources.

In addition, the team built two localized LLM-driven knowledge centers, featuring a centralized knowledge base for all users and a tenant-based knowledge base for private queries. Now, when engineers are attempting to resolve network issues, they can use natural language intent to search for solutions – for example, “We have a fault on this transmission link with alarms 7750 and 7401; what do these alarms mean and what is the best way to fix the fault?” or “The KPI for throughput has dropped below threshold; what can I do to optimize performance and improve the KPI?”

China Mobile is rolling out the GenAI solution in three phases. The first, called Intelligent Assistant, was completed in 2023 and helped the operator achieve Level 3 autonomous networks. The results detailed in this case study were achieved in the Intelligent Assistant phase.

The second phase, called Expert, is happening now and will be complete in 2025. During the Expert phase the team is building on LLM to automatically create network configuration instructions and optimization directives in complex operations scenarios and implement solution recommendations. China Mobile hopes to reach Level 4 autonomous networks as a result. The final Master phase in 2025 and beyond will support more autonomous end-to-end operational processes with the aim of achieving Level 5 autonomous networks.

Measurable impacts

The measurable results so far of the GenAI solution have been significant, with an overall savings of $7 million:

• Natural language dialog boxes enable engineers to quickly acquire the knowledge they need, with most questions answered comprehensively within one minute. Overall, this has resulted in a 67% increase in knowledge coverage.
• Developers can now use natural language for data development and exploration and to generate intelligent data analysis reports. This has resulted in $2.9 million in annual cost savings.
• A natural language interface supporting multiple data sources automatically generates and executes SQL queries to retrieve data, create task schedules, perform data processing and generate reports. As a result, front-line engineers have reported 72% faster data analysis on average, equating to a total of 13,545 workdays saved and $2.5 million in opex savings annually.
• The time it takes to create operation reports has been reduced to an average of one day. With 120 engineers each developing four reports per month, this has resulted in annual savings of 11,520 workdays and $1.6 million.
• Time needed for feature development has been cut to an average of five days. With 80 engineers producing an average of 2 features per person-month, this has led to an annual savings of 7,680 person-days.