Agentic AI success in telecom depends on the availability of meticulously curated, multi-domain data derived from network transactions and interactions
Agentic AI tops many telecom providers’ strategic objectives for 2025. Recognizing its potential, some providers are already investing in AI agents to enhance customer care. This initial step is followed by a more comprehensive evaluation of advanced applications for network operations
Investments in AI are crucial for telecommunications operators as they aim to reduce operational costs, improve efficiency and enhance customer satisfaction. These challenges are exacerbated by the urgent need to modernize legacy infrastructure to align with evolving network standards, such as 3GPP Release 18 or newer generations of technology like 5G-Advanced.
Agentic AI can effectively address these challenges and unlock significant value for both customer experience and network operations. However, its success depends on the availability of meticulously curated, multi-domain data (I.e., RAN, CORE, MEC and Transport) derived from network transactions, interactions and the understanding of the relationships between each network function within the mobile network architecture.
Customer experience: Driving efficiency and personalization
AI agents are already revolutionizing customer experience across the industry, directed toward improving operators’ Net Promoter Score (NPS) and Average Revenue Per User (ARPU), while reducing customer churn. Here are two key areas where agentic AI is poised to significantly improve the customer experience:
Intelligent Virtual Assistants:
Unlike traditional chatbots that answer preprogrammed questions, agentic AI bots comprehend intricate queries and can independently resolve issues or complete transactions via integrations with billing, tech support and sales. They also provide 24/7 support, leading to faster resolutions to common issues like setting up a new device or billing inquiries.
Personalization:
As agentic AI advances, bots can personalize service based on customer usage patterns, behaviors and preferences. This technology would allow operators to anticipate needs and present proactive offers or upgrades.
These capabilities transform customer service from a traditional cost center into a strategic function that actively protects revenue and potentially builds long-term customer loyalty.
Network operations: The next frontier for agentic AI
While relatively few providers are nearing full integration of AI agents into network operations, as agentic AI gains broader adoption, the technology stands to fundamentally transform operations, making them more autonomous, efficient and secure. With that said, here are two areas AI will make the most significant impact on network operations:
Predictive Maintenance and Proactive Assurance:
As operators progress in their technological advancements, agentic AI will possess the capability to analyze vast quantities of real-time network data. By leveraging historical performance data, it will be able to predict potential failures before they occur. This proactive monitoring will enable operators to identify service disruptions early on, allowing them to transition from reactive repair to proactive maintenance strategies.
Intelligent Network Security and Threat Response:
Upon full integration, agentic AI would continuously monitor for cyber threats and anomalies, detect suspicious activities and proactively implement countermeasures, such as blocking malicious traffic or isolating compromised devices. Leveraging machine learning, agentic bots adapt to new threats and expedite the time required to stop a threat or alert personnel for human intervention.
These use cases will have a profound impact on companies, but the benefits will only be realized if the AI is trained using curated, diverse domain data.
The crucial role of multi-domain data
For AI to deliver optimal outcomes, operators need two types of data. First, they require real-time packet-based data to gain visibility into their mobile network infrastructure. Second, they need historical data, including performance metrics, logs and incidents, which provides essential context to understand trends and patterns. This data enables the development of accurate predictive analytics and automation. When combined, these data sets facilitate proactive troubleshooting and the prediction of likely events.
To obtain a comprehensive view of the entire network, operators must collect data from various sources. This includes real-time Trace Port (RAN), deep application-level and latency visibility (MEC) and Control Plane/User Plane analysis via deep packet inspection (DPI) for baselining application behavior. DPI sensors inspect the actual data payload of traffic transversing the network. DPI data analyzes network traffic, troubleshoots subscriber performance, and, when enriched with Control Plane metadata, provides subscriber-level context. This unique capability of DPI sets it apart from other network monitoring methods, enabling organizations to gain insights into the contents of traffic and train AI systems that require a precise understanding of network behavior.
Crucially, data intended for training AI agents must have the appropriate context and level of enrichment to provide data quality with context, rather than data for its own sake. We’ve all witnessed the world of big data, where vast amounts of network data are loaded into a data lake, and data scientists are expected to conjure magic. However, the carriers have learned some lessons and are now transitioning to more refined “curated data,” streamlining the entire AIOps pipeline. This advancement leads to more precise results, reduced compute resources and decreased data lake storage for redundant data. Instead of simply producing data, curated data must be produced, ensuring high-quality, accurate and relevant information crucial for effective AIOps.
Furthermore, telecom knowledge is vital for contextualizing multi-domain data into meaningful, actionable intelligence. For instance, it can enrich data with absolute identifiers like International Mobile Subscriber Identity or Subscription Permanent Identifier, even if it was never encountered on the network. Achieving this context, however, demands domain-level knowledge and expertise to effectively enrich raw data. Companies often encounter logical hallucinations when they use less granular, accurate and clean data to train their AI models. This issue is detrimental because it necessitates human intervention to investigate and resolve, diverting time and resources away from more critical matters.
The need for human expertise
Even with the right data to train AI, human intervention is still necessary to govern it. Engineers have built up telecommunications networks over decades of experience. When outages occur, sometimes it takes an investigation to realize that, until it fails, a non-redundant part of the network was impeding traffic. This expertise is crucial to comprehend the underlying reasons for a malfunction, which hinders an AI agent from resolving an issue without creating new ones.
Experts are essential to curate the data initially and then collect, clean, validate and label it for the training models. Humans must also design AI models and train agentic agents with precise, high-quality feedback to correct mistakes. This establishes ethical guardrails and protocols that the AI must follow. Only then will AI be able to fully understand what to do in different situations.
With nearly all telecommunications operators who Nvidia surveyed for its State of AI Telecommunications: 2025 Trends report saying they are adopting or assessing AI, agentic AI’s growth will come. While intelligent virtual assistants are already providing 24/7 support and personalizing interactions, broad implementation for network operations for predictive maintenance and proactive cyber monitoring is further away. Operators who commit to using high-quality, domain-specific data will be best positioned to capitalize on this future, ensuring agentic AI becomes a powerful tool across the industry.
