Powering the future of telecom: Cloud, AI and the rise of autonomous networks

Eoin Coughlan, Global CTO Telco, Media and Entertainment Industry, IBM Technology, will be sharing a fireside chat about AI in Telecom: From Early Wins to Scaling Success – Real and Actionable Insights with Haithem Alfaraj, Group CTO at stc, on 18 June at 11.55, Stage C. Here he talks to TM Forum about the critical shift to autonomous networks (AN) operations, overcoming the limitations of Gen AI, scaling it and making AI network-aware.

TM Forum: Why is the shift to fully AN operations a critical milestone for telecom providers? How does a purpose-built approach overcome the limitations of traditional Gen AI models in delivering true network automation?

EC: In a recent survey by IBM and TM Forum, 69% of executives recognize that intelligent orchestration and programmability are essential for superior customer experiences, while 63% say ANs can enable new service models and drive revenue growth. The rapid growth of 5G, IoT, and edge computing – combined with the shift to virtual and cloud-native solutions – has made network environments increasingly complex and difficult to manage. Traditional, manual or rule-based systems struggle in this dynamic landscape.

ANs, powered by AI and purpose-built Gen AI, enable self-management, self-healing and dynamic adaptation – transforming operations from reactive to proactive. This shift results in improved service reliability and customer experience, reduced operational costs and more efficient scalability.

Traditional Gen AI models often lack the deep domain-specific knowledge and real-time responsiveness required in telecom environments. In contrast, purpose-built, ‘verticalized’ Gen AI solutions are infused with telco-specific data, context, and operational logic at every layer.

This tailored approach enables real-time decision-making, seamless automation across network domains, and the ability to manage complex, multi-vendor environments – delivering true end-to-end network automation that generic models cannot achieve. Importantly, small language models, when fine-tuned, and time series models provide the efficiency and performance necessary for accuracy and return on investment (ROI).

My colleague Andrew Coward, who is GM, Software Defined Networking at IBM, will be co-presenting a session called Building Trust in Autonomous Networking with Jared du Plessis, Cloud Engineering Manager at Vodafone, on 17 June at 12:00, Stage B.

TM Forum: How can telecom providers ensure predictive, network-aware AI models are both accurate and trustworthy, balancing sensitivity with precision while relying on the right data foundations and governance frameworks?

EC: Although 80% of collected data reportedly goes unused, achieving business value hinges on acquiring the right data – with the appropriate granularity, latency, frequency and reliability – to support targeted use cases. In addition, networks require optimal models –such as time series models and large language models – with the flexibility for continuous optimization and replacement as technology evolves.

This demands robust data foundations that integrate network telemetry, customer data and operational logs, along with governance frameworks that ensure data quality, traceability and transparency. Explainable AI and regular audits are essential to validate decisions and correct biases. The selected AI platform, or platforms, must effectively manage proprietary, self-created and open-source models.

As model superiority evolves rapidly, continuous refinement using historical and real-time data is critical. Calibrated, auto-generated alert thresholds that are focused on high-impact applications like predictive maintenance and network optimization help to maximize accuracy and business value. Data and AI must be managed, open, flexible and governed.

TM Forum: Many service providers operate with siloed systems or vendor-specific AI tools. What does it take to scale Agentic AI across multi-vendor, multi-domain network environments to unlock productivity across all departments truly?

EC: To scale Agentic AI effectively across multi-vendor, multi-domain telecoms environments, organizations must first focus on unified data integration. This includes consolidating disparate data sources into a single AI-powered platform that enables cross-domain analytics and decision-making through standardized APIs, data lakes and integration layers – preserving contextual relationships while normalizing data from diverse systems.

Success requires vendor-agnostic AI frameworks. CSPs must build abstraction layers that allow AI agents to interact seamlessly across vendor systems using consistent interfaces, reducing proprietary lock-in. TM Forum APIs that support this approach include: TMF632 Party Management API; TMF639 Resource Inventory Management API; and TMF638 Service Inventory API for the core infrastructure.

The Forum’s APIs for operations are TMF656 Service Problem Management API; TMF628 Performance Management API; and TMF641 Service Ordering API.

For the orchestration layer, robust workflow automation is essential to coordinate AI agents across domains for complex, multi-system tasks. This includes deploying orchestrator agents that manage specialized sub-agents across network management, customer support and billing. Over time, it is vital to implement continuous learning capabilities, enabling agents to improve accuracy, performance and efficiency through operational experience.

To drive effective multi-vendor AI integration, organizations must use orchestration capabilities based on common taxonomies and semantic models. This allows AI agents to share understanding across operational boundaries while maintaining context. A comprehensive governance framework must also be established, with clear oversight mechanisms addressing security, ethics, and compliance uniformly across all AI deployments – ensuring consistent management regardless of vendor origin.

This comprehensive approach dismantles organizational silos, achieves end-to-end automation, and drives significant productivity improvements across all functional areas.

At DTW, we will present a preview of our upcoming joint study with TM Forum, Modernizing Telco Operations – Leveraging Cloud and AI for Network Automation, which reveals that organizations operating at Level 3 and 4 AN levels see a 31% higher ROI from their investments compared to those at Level 1 and 2, while investing only 17% more.

Join IBM at Booth #320, Hall C3, to know more.