Preventing hidden network risks before they become outages

Finding risks hidden behind network change

Communications service provider platforms generate millions of raw alarms every day. Automatic filtering and compression can make those volumes manageable, but the process can also remove context that would reveal an emerging risk. Hidden issues can then develop into service outages before operations teams understand their significance.

Frequent network and service changes add another challenge. Live topology moves further from its original design over time, and each change can introduce new blind spots or single points of failure. Continuously tracing these updates and assessing their service impact is beyond the capacity of manual investigation alone, leaving network operations teams in reactive firefighting mode.

The Catalyst project, Agentic NOC: Model-driven risk prevention at scale, aims to identify and resolve potential risks before customers are affected. It focuses on detecting risk created by network drift, prioritizing findings through service-impact analysis and generating a response that can be executed remotely or supported through an on-site work order.

Moving from reactive restoration to preventive operations

The solution is designed around intent, awareness, analysis, decision-making and execution. Full-scale alarm analysis is combined with a risk agent, digital twin networks and domain model to retain critical context. A risk agent collaborates with resource agents to identify potential issues, assess their likely business impact and recommend the next action.

Deep learning trained on historical network behavior and fault patterns provides the analytical foundation. The EDNS algorithm maps events such as link interruptions and performance degradation to key business indicators, including affected users, traffic loss and SLA-breach risk. A spatial-temporal fusion transformer architecture infers future network behavior in KPI values and simulate service impact.

This gives operations teams a basis for prioritizing recovery resources according to service impact rather than alarm volume alone. The solution can generate an appropriate response and execute it remotely, or dispatch a field maintenance engineer with work instructions delivered through the work order and supported by Copilot.

The approach also changes the operating model. As the balance between people and machines evolves, the project identifies a need for new roles, responsibilities and skills. Human expertise remains part of the process, but it is redirected from repetitive alarm handling toward reviewing evidence, managing exceptions and resolving higher-value operational risks.

TM Forum assets shape the agentic operating model

The project uses the AI Native Agent Factory as an architectural reference. GB1087 informs agent security, while IG1505A supports business indicators, Autonomous Networks processes and an ODA-aligned solution architecture.

GB1041 defines the value tree used to quantify the solution’s contribution. TMFC041 Anomaly Management, TMFC043 Fault Management and TMFC038 Resource Performance Management support risk detection, fault handling and resource monitoring.

TMFC061 Work Order Management supports operational execution, while TMFC014 Location Management supplies equipment and personnel location information.

GB1505A is used to assess Champions’ Autonomous Networks maturity for mobile broadband fault management, while GB1505B is being contributed for home broadband problem management. Together, these assets provide a consistent foundation for preventive, agentic network operations across different service domains.

Targeting fewer outages and lower emergency costs

The project targets a potential 20% reduction in major service outages by identifying risks before they become customer-impacting faults. The intended operator benefits include stronger network resilience, preserved SLA compliance, fewer customer impacts and lower ad hoc firefighting and emergency maintenance costs.

The commercial case is closely connected to service continuity. Undetected risks can escalate into outages, exposing CSPs to SLA penalties, damage to customer experience and brand reputation, and higher emergency repair expenditure. A preventive operating model aims to reduce those impacts by turning alarm and topology data into prioritized, actionable risk intelligence.

“By proactively identifying hidden network risks, this solution can strengthen resilience and improve network operations center efficiency. Deep learning, digital-twin networks validation and domain models can uncover risks that engineers may not easily detect,” said Boonchoung Tansuthepverawongse, Head of Network Data and OSS Management at AIS (Advanced Info Service).

A reference model for preventive agentic operations

For the wider industry, the Catalyst aims to establish an agentic operations reference model for detecting risk, simulating impact and preventing disruption. It addresses the loss of context caused by alarm aggregation and the blind spots introduced by frequent topology change, providing a reusable approach for converting large volumes of alarm and topology data into risk mitigation.

For society, more stable communications can support continuity for consumers, enterprises and home services. By moving operations from reactive incident restoration toward proactive risk prevention, the project offers a route to more resilient infrastructure and fewer unplanned disruptions.

Meet the team at Innovate Asia in Bangkok, 17-19 November 2026.