ANL4 deployment at scale is limited until it can be trusted

Building trust in AI-native network operations

Autonomous Network Level 4 represents a major leap in telecom operations, enabling networks to self-monitor, self-heal and optimize in real time. But for CSPs, the move from automation to autonomy depends on whether Agentic Operations can be trusted to make accurate, secure and controlled decisions in high-impact network environments.

The Catalyst, Building trust in ANL4, addresses one of the core barriers to Level 4 adoption: confidence in AI-driven operations. It focuses on multi-agent support across domains and vendors for service outages, service degradation and risk prevention, where hallucination, incorrect API execution or actions outside defined guardrails can directly affect service performance and customer experience.

The project supports the safe rollout of AI-driven “Self-X” network capabilities, helping operators intervene before risks become incidents major outages and enabling a maximum one-hop closed loop during service outage and service degradation. It also aims to move operations from reactive response to predictive stability, improving service availability while reducing pressure on human operations teams.

A 3R approach to trustworthy agentic operations

The Catalyst introduces a multi-agent architecture built around a “3R” approach: robust data governance, rigorously trained large language models and retrieval quality and precision.

At the data layer, the team strengthens governance and guardrails for the Digital Twin Network using TM Forum standards, ensuring data quality, integrity and compliance across the autonomous operations lifecycle. This creates a trusted foundation for AI decision-making.

Building on this data foundation, the system applies domain-specific telco models and incrementally trained LLMs, using standard operating procedures alongside foundation models such as DeepSeek and Qianwen. This helps ensure that AI decisions are context-aware, aligned with telecom operations and less prone to hallucination.

To achieve end-to-end autonomous execution, advanced retrieval-augmented generation (RAG) drives knowledge collaboration and precise data retrieval, while DTN Graph AI models complex multi-vendor, cross-domain relationships. This is enhanced by chain-of-thought reasoning and backtracking supervision to improve accuracy and make decisions inspectable. Together with multi-agent orchestration, these capabilities automate the incident lifecycle, from proactive monitoring and fault isolation to root cause analysis and task dispatching.

The project uses and contributes to a broad set of TM Forum assets, including GB1505A Individual Service Fault Management Questionnaire, IG1505A Individual Service Fault Management Solution Package, IG1453 Agent to Agent Protocol for Telecoms, IG1412 AI Agent Specification, IG1414 Agentic AI Closed Loop, IG1301 Measuring and Managing Autonomy, IG1291 MAMA CSP Value Streams for Autonomous Operations, GB1041 Value Operations Framework, GB1042 Autonomous Operations Maturity Model, GB1079B DT4DI for Service Management, and Open APIs including TMF642 Alarm Management, TMF656 Service Problem Management, TMF649 Performance Thresholding Management and TMF921 Intent Management.

Operational impact from real-world AN L4 deployment

The Catalyst is grounded in deployment by Indosat Ooredoo Hutchison (IOH), where agentic fault management has been applied across wireless, microwave and IP domains, which together account for 80% of overall network incidents.

The reported operational impact is significant. IOH has improved end-to-end fault MTTR and performance degradation MTTR by 20%, reduced major incidents through risk identification and prevention by 20%, and achieved a 30% increase in NOC efficiency. The deployment has also supported more than 60 digital staff, enabled more than 40 AI Builder, AIOps and DataOps taskforces, and expanded human and AI co-working coverage from 20% to more than 80%.

Customer and business outcomes are equally strong, including a 56% reduction in loss hours, a 19% reduction in total incidents, around 20% lower service loss and a 62% reduction in network-related churn.

As Luthfi Auzan, VP, Head of Operations Transformation & Analytics, Indosat Ooredoo Hutchison, explains, “The project delivers clear tangible benefits: supporting business outcomes including ARPU uplift by around 10%, traffic increased by 29%, latency of key apps reduced by around 50%, B2C complaints reduced by 52%, B2B complaints reduced by 32%, talent transformation of 20% of headcounts, and operations assurance and insight value of more than $10 million a year.” He adds that the Catalyst also provides “a proven, scalable blueprint for the global telecom industry to adopt AN L4 autonomous operations,” supported by IOH’s contributions to shared TM Forum industry assets and reference resources.