What does it really take to make AI useful in telco operations? That question sits at the heart of The Mobile Network’s latest market report, The AI-Native Telco. The following article originally appeared in the report.
1. Meeting operators’ historic challenge
Operators want to harness the benefits of AI and automated operations, but face challenges due to the accumulation of legacy tooling and siloed data. This has led to data being scattered across diverse business departments, geographies and systems. Data can be stuck in spreadsheets or other non-queryable formats, or it could be that different business teams use non-compatible tools and systems.
All this contributes to the fact that data is not unified, and is not being shared – or is not currently able to be shared – between teams.
For example, customer-facing APIs that allow end users to check their network coverage are typically embedded within customer care workflows. Yet the same data could provide network operations teams with an early warning system into a potential site outage or localised service issue where there is a spike in the number of customers running service checks.
Equally, real-world performance data is key for RAN planning teams to understand and optimise the network effectively. Yet this data could be connected back into customer care, enabling operators to proactively update customers about an issue on the network before they contact the support centre.
In many cases, the data already exists. The challenge is connecting it across the organisation so its full value can be realised.
2. Being in a prime position to build AI solutions
Over nearly two decades, Metricell has built a broad portfolio of software and services, underpinned by deep expertise in telecoms operations.
Our solutions ingest data from a wide range of sources including performance management tools, network OSS and crowd-sourced measurement data, as well as planning software, business systems, third party and MDT (Minimisation of Drive Test) data. Currently our aggregated data-based solutions are accessed via browser-based platforms, which are designed to make complex data usable for operators.
Having solved the data access and correlation problem we are in a prime position to build on this experience to integrate AI to our solutions, and we are now building Agentic AI use cases on top of this consolidated network and business data.
3. What “Understanding the Data” Really Means
To do that means truly understanding the data that AI agents will be accessing. That means knowing which data sets matter to specific internal users, from radio and network engineers to customer care, business and enterprise account teams. This involves presenting the same raw data with different levels of detail, language and actionability.
The outcome is to create a shared source of data for multiple persona-specific views and explanations. For example, a customer support agent would have access to the same deep network data that engineers use, but that data is translated into a clear explanation for the agent as to what is wrong and how long it might take to fix, providing simple reasoning the agent can then communicate to the customer.
The AI-native operator cannot have one intelligence layer for engineers and a separate one for customer facing teams. Data must be integrated to create a true digital twin of the network, with human teams interacting where appropriate via a single plane of glass.
4. Agentic AI: The New Layer on Top of the Data
Our AI engineers are developing AI agents with access to this data layer, giving them the ability to reason, investigate, and propose actions. The important factor here is the deployment of well-designed tools and AI skills. Telco data sets are huge and deeply interlinked. Any company can call public frontier models but running all your telco data through a generic model does not deliver benefits.
Metricell links models to proprietary telco data, understanding that data and the operator’s workflows. This develops a clean, correlated, API-accessible layer that sits over heterogeneous telco data. It prepares data explicitly for AI consumption, not just human dashboards.
“AI will only be as effective as the OSS/BSS environment it operates within. Many operators are already managing fragmented systems, inconsistent data models and disconnected processes. Adding AI without addressing these underlying challenges simply amplifies complexity.
To maximise the value of AI, operators need to rationalise their OSS/BSS landscape and prioritise data quality, governance, and integration. They should work with partners who combine deep telecoms expertise with advanced technology capabilities, ensuring AI solutions are built around the realities of network and business operations.
By designing domain-specific tools and providing controlled access to rich, interconnected datasets, these partners enable AI agents to deliver reliable, scalable, and operationally relevant outcomes.”
- Francisco Pareja
5. Empowering your engineers
One important element of the AI-Native telco is that it will actually give the telco more control over their direction and operations. We see engineers being able to handle multiple tasks in parallel, becoming effectively skill authors for AI – moving from operating tools and opening tickets to configuring and refining AI skills that perform those tasks.
By tuning and governing skills that capture operator-specific processes and reflect local priorities, thresholds and policies, telcos will be able to continue to differentiate themselves it the market.
Being an AI-Native operator is not about accessing generic, one-size-fits-all agents. Operators need an AI path that embodies their own playbook. That way AI becomes a custom asset, not a commodity tool.
In the longer term, as AI Agents directly open tickets, schedule work and apply approved network changes, the human role shifts to defining guardrails and approving and monitoring automation policies.
6. A practical approach
Currently we see that Agentic AI is in a pilot phase with operators, with a focus on high-value, “step-one” use cases that can be integrated into existing platforms.
One example of this is a feature we have developed which uses an LLM to ingest complex trouble tickets and related data sources to output tailored summaries for different teams. The key benefit of this use case is the time saved – what used to take many minutes can be reduced to a few seconds. Rather than searching through multiple documents to find the right one, or reading through pages of complex engineering terms, the feature does this automatically and presents the key information back to a customer care agent or operations manager in a plain language output. Features like this are essential for demonstrating success and building strong business cases to support future AI projects.
Operators have a natural reluctance, or are simply not allowed, to send sensitive network and customer data to external frontier models. Because of this need for control, privacy and compliance, we provide GPU infrastructure and data centres where operators can host their own models, keeping data sovereign within controlled facilities.
These hosted models are approaching the capability of frontier models from a year or two ago, which is increasingly sufficient for many telco use cases.
“In the short term, the industry should create pathways that allow teams to trust the technology, as this is also a change management process. Developing AI tools which are genuinely useful, and integrating them into existing platforms today, builds trust with teams on the ground and delivers value for operators in the short term.
This approach underpins the longer term goal of moving to full, AI-native platforms that are trusted from the ground up. It’s a new way of thinking, so teams need to be supported through the transition to AI or these platforms will be rejected, even if they bring significant benefits.”
- Luke Alexander
7. Data-First AI as Telcos’ Real Competitive Edge
Gaining the advantage as an AI-Native operator is about establishing data and domain understanding, building trust, skills and managing change. AI-native operations are continuous, providing always-on intelligence. That requires regular model tuning, data updates, workflow refinement and managed service support. Metricell is a vendor and model-agnostic data aggregator with that capability, developing solutions that are now layered with Agentic AI.
Our agents and use cases respect the complexity of telco data and are built to be operator-specific and engineer-led. We know that operators that master the use of AI - from engineering to customer care to business insight - will be better positioned to reduce operational friction, differentiate customer experience and move confidently toward more autonomous operations and the AI-Native future.
