Improving Service Quality with AI-Powered End-to-End Automation

Each year, TM Forum’s Digital Transformation World (DTW) Ignite event highlights tangible efforts by vendors and operators to modernize the telecom industry. Collaboration is always at the heart of the show, and with numerous catalysts, this year was no different.

TM Forum’s highest-profile initiative is its Autonomous Networks mission and the collaborations around a standard open architecture to enable zero-wait, zero-touch and hopefully zero-trouble operations. AI has long been a part of this effort, but the importance of generative AI is rising because of its ability to tie together unstructured information and enable an end-to-end approach.

Using AI, ZTE demonstrated further progress with its AIR Net autonomous network solution at DTW in Copenhagen this year, building on the announcement at MWC 2025. The extensive use of AI includes large language models (LLMs), with which ZTE aims to help operators improve efficiency and reduce costs. With rising network complexity, autonomous network operations are becoming increasingly important. AIR Net focuses on three areas: infrastructure, single-domain network operations and cross-domain service operations, covering over 20 important autonomous scenarios.

Why Autonomous Networks Matter to Operators

The top three drivers of autonomous network deployments revealed in a recent TM Forum operator survey in June 2025 align closely with ZTE’s approach. The top reason was to improve customer experience or satisfaction, cited by 90% of respondents as very important. Cost reduction came in second, selected by 85% of operators, followed closely by simplification and improving personnel efficiency, with 84% rating it as very important.

The name AIR Net comes from “AI Reshaped Network”, reflecting the magnitude of the changes that are needed in operations and maintenance. AIR Net uses small and large models in tandem, alongside digital twins, to boost the skills and knowledge of network team members across previously siloed domains. Over time, ZTE expects this will gradually replace manual tasks with fully autonomous closed-loop automation.

Saving costs by reducing human involvement isn’t the only operational cost-saving. There’s also the potential to reduce network energy needs and boost profitability through faster time to market or to expose network-as-a-service capabilities, for example, through Camara network APIs. Among operators, 52% cited green energy savings as very important and a further 42% said they’re somewhat important. Although it ranked lower as a benefit, fault management and anomaly detection was the top process that operators are addressing first, with 82% saying it’s a high priority.

Agentic AI Moves to the Centre of the Stage

To enable end-to-end automation, systems must communicate directly with each other. ZTE is enabling AI agent-to-agent interaction to deliver this goal. In April 2025, ZTE open-sourced its Co-Sight Framework, which separates analysis from execution. There’s an increasing number of agentic approaches from different AI-focused companies, such as OpenAI or DeepSeek, which is why Co-Sight supports standard protocols for cross-vendor LLM agentic connectivity, for example, Anthropic’s Model Context Protocol or Google’s Agent2Agent protocol. This enables greater flexibility for operators to choose the best AI model, or models, for their needs.

ZTE also has its own large model, the Nebula Telecom Large Model, which is pre-tuned specifically to telecom network characteristics and tasks. This is one of the three core parts of the AIR Net solution, alongside a data engine and digital twin engine. But because the approach is modular, there’s flexibility for operators that deploy the system.

Six different agents are intended to interact as part of AIR Net. At each step, they report their analysis and the proposed actions in natural language to enable human oversight. In essence, this is akin to having a safety driver in an autonomous vehicle: it ensures the agents successfully accelerate resolution and operations, and a human overseer makes sure that no unexpected new issues are created by the AI agents.

Each agent attends to a specific step in the process: identification, pre-processing, delimitation and location, impact analysis, scheduling and evaluation. In essence, the large telecom model is trained on each task to improve performance and ensure it fully understands potential problems and the actions that can be resolved. The systems include an optional mobile interface for human agents to interact with the agents.

Collaborations on Show at DTW

Building telecom networks is a team game: many companies must coordinate their product offerings. In the same way autonomous agents need to interact with each other, so do different communications service providers and a range of vendors. The results that operators see from cross-domain automation are worth the implementation efforts.

Operators report significant improvements in key operational metrics. In Shandong, China, an operator showed a collaborative cross-domain agent that worked across IP, fixed and mobile network event management and improved the mean time to resolve by 10%. Also, it boosted the accuracy of root cause analysis to over 90%.

An operator in Jiangsu reports an improvement of 15% in the average time to recover from target faults and a 12.7% reduction in energy usage per gNodeB or 5G base station. In Hunan, an operator has seen the average complaint handling time reduced by 10%. Perhaps most impressively, in Henan, a Chinese operator improved the time to locate root causes of problems from four hours to 10 minutes.

In Shandong, an operator has used ZTE’s agents to streamline customer complaint handling across domains. Identifying the part of the network that originates the issue, the demarcation, is handled by an agent. Similarly, in the problem domain, there’s a closed-loop system for handling and optimizing the solution. The operator reports that 90% of cross-domain problems are now automatically demarcated, with an accuracy rate of 85%, and the pre-processing time to demarcate issues has been reduced by 60%.

These systems demonstrate that operators can improve efficiency using automation even at the current stage of autonomous networks. On TM Forum’s five-point scale, the leading operators are generally still scoring less than 4, indicating there are more gains to be had. One Chinese operator scores 3.8 on wireless fault scenarios and 3.6 on core network fault scenarios. As operators progress further with autonomous network automation, there should be further improvements in the speed of innovation, cost savings, energy usage, fault resolution speed and overall efficiency. This should lead to improved customer satisfaction.

Written by:
Posted on July 22, 2025
Share