Raja Shah, Executive Vice President, Industry Head – Global Markets at Infosys, talks about the opportunities and challenges of operationalizing multi-mode AI at scale, including getting the foundational data architecture right. Providing accurate, accessible and interoperable data is the key, he says.
Sponsored by:
Unlocking the power of data and AI is a strategic Imperative for telecom
As the telecom industry undergoes rapid digital transformation, the convergence of data and AI is emerging as a cornerstone for innovation, efficiency, and customer-centricity. For communication service providers (CSPs), the ability to harness data and deploy AI at scale is no longer optional – it’s a strategic imperative, says Raja Shah (RS), Executive Vice President, Industry Head – Global Markets at Infosys in this interview with TM Forum.
RS: Telecom operators sit on a goldmine of data such as network telemetry, about customers’ behavior, usage patterns and more. For example, AI enables CSPs to transform this raw data into actionable insights such as hyper-personalized, omnichannel experiences by analyzing customers’ behavior in real time and automating contextual responses.
Increasingly, agentic AI systems – which can autonomously make decisions and take actions – are enhancing these capabilities. For example, agentic AI assistants can proactively resolve customer issues, escalate complex cases, or even initiate service upgrades based on predictive insights, all without human intervention.
In many use cases, GenAI-powered assistants help reduce resolution times and improve satisfaction. AI-powered chatbots and virtual assistants are reducing call center loads while improving customer satisfaction. Meanwhile, predictive analytics are helping operators anticipate network failures before they impact services.
RS: Despite the promise, many telcos face significant hurdles in scaling AI because they are hampered by the quality of their data and the fact it is held in silos. A modern, AI-native data architecture is foundational – TM Forum’s AI & Data mission emphasizes the need for accurate, accessible and interoperable data to train AI models effectively.
Other challenges include legacy IT which impedes agility, and CSPs lack AI talent and data science capabilities. There are also important regulatory and ethical concerns to address around data usage.
To overcome these challenges, telcos must adopt a platform-centric approach that embeds AI, and increasingly agentic capabilities, into core business and network operations. Our participation in TM Forum’s Catalyst projects, such as Unleash the potential of GenAI-powered 5G network slicing and Agent Fabric – Phase II demonstrate how AI can be scaled using the Forum’s Open APIs and Open Digital Architecture principles. Success hinges on aligning leadership, standardizing data architectures and leveraging reusable AI components to reduce time-to-value and operational risk.
Leading CSPs are already investing in cloud-native platforms, adopting MLOps practices and building cross-functional AI teams that bridge business and technology. In general, the industry is moving toward a unified blueprint that includes ontology, intent-based automation and standardized APIs to support scalable AI deployments. This ensures telcos can evolve from digital to AI-native operations.
The role of data governance is also essential in operationalizing AI because trustworthy AI begins with robust data governance so telcos must ensure that data is: accurate and complete; secure and compliant with regulations like GDPR; and ethically sourced and used.
To address privacy and compliance concerns, federated learning is gaining traction. It allows CSPs to train AI models across decentralized data sources – such as edge devices or partner networks – without moving the data, ensuring GDPR compliance and data sovereignty. Frameworks such as TM Forum’s AI Governance Toolkit provide valuable guidance on managing AI risks, ensuring transparency and embedding accountability into AI systems.
Network intelligence is one of the most impactful applications of AI in telecom because the models can produce such a range of insights. They include predicting traffic surges and dynamically allocating bandwidth, detecting anomalies and preventing outages.
Another important benefit is that AI can optimize energy consumption across network assets. Technologies like Self-Organizing Networks (SON) and AI-driven optimization of the RAN are already delivering measurable improvements in performance and cost savings.
CSPs have much to gain from AI-enabled predictive maintenance and network optimization, fraud detection and revenue assurance, and automated operations that drive cost efficiencies. Causal AI enhances this by identifying not just correlations but true cause-and-effect relationships which are critical for root cause analysis and reliable decision-making in dynamic environments.
Agentic AI enables self-healing networks – systems that detect, diagnose and resolve issues autonomously. These agents can coordinate across domains, optimize energy consumption, and even reconfigure network topologies in response to real-time demands.
RS: By 2030, telcos will be autonomous, intent-driven and platform-based. They will operate zero-touch networks, deliver AI-curated services and monetize data through B2B2X ecosystems. TM Forum’s vision includes embedding AI in every layer – from network slicing to customer engagement – supported by ethical frameworks and industry-wide collaboration.
We are looking ahead to a data-driven future because as 5G, IoT and edge computing continue to expand, the volume and velocity of telecom data will skyrocket. The winners in this new era will be those who can unify their data ecosystems and embed multi-modal AI intelligence – from generative to causal and agentic AI – into every layer of their operations. Telcos must also continuously learn and adapt through intelligent automation.
By embracing a data-first, AI-enabled mindset, telcos can improve operational efficiency, unlock new revenue streams and deliver exceptional customer value.