AI output is only as good as the data it feeds on
Integrating AI within your network end-to-end, from IT to business, will be key to realizing the agentic future telcos are striving for
In my conversations with operators, I'm seeing serious and growing ambitions around AI. Building an agentic workforce to work alongside human teams is becoming a strategic priority. These agents are designed to sit inside the network, working together to query data and act in real time.
The potential is obvious: Better customer experience, Greater operational efficiency. Faster decision cycles. More responsive networks, from customer care through to network utilization. Core systems such as BSS and OSS will increasingly be operated by AI agents, and processes like application modernization, cloud migration and testing will move in the same direction.
But there's a serious data problem
The vast majority of data is unstructured, and it can't actually be used in its raw state by the autonomous AI-driven workflows operators are aiming to build. There's real autonomy risk when AI entities are entrusted with decision-making based on inconsistent or incomplete data, and that's because AI agents are only as good as the data they incorporate, and the context they understand. Operator data environments create a challenge for agentic AI as it needs to access data in real-time from across a heavily siloed, multi-system, multi-protocol, multi-cloud technology landscape.
My experience in telecom – particularly working at the intersection of AI and data – has taught me that preparing data for AI isn't just about building better models or experimenting with new algorithms. The real work lies in structuring, aligning and governing data so AI systems can operate reliably in production.
Customizing data strategies and preparing telco data for use by purpose-built AI frameworks is one task that AI can't perform, but disciplined data services can – that's why they play such a critical role in determining whether risk-free autonomy is achievable.
"Disciplined data services play a critical role in determining whether autonomy is achievable or not."
To successfully implement AI and embed intelligence across all workflows, it comes down to domain expertise.
The only way you can get structured, controlled results from AI is to provide it with an end-to-end understanding of the taxonomy of the network. This requires AI agents to be designed and built upon a domain-focused intelligence foundation.
Autonomy exposes the limits of today's data foundations
In my conversations with telco operators about autonomy, I'm seeing the same pattern: AI strategies are advancing faster than the data foundations beneath them. Agent-driven workflows aren't copilots. They execute bounded decisions inside production systems, and at that level, their tolerance for fragmented or inconsistent data drops sharply. Most telco data environments weren't designed for real-time, cross-domain execution, and decades of layered BSS and OSS systems have created structural fragmentation:
- Network events in one domain
- Customer context in another
- Billing hierarchies elsewhere
- Service inventory in yet another system
Even when the data exists, it isn't consistently correlated or synchronized.
Autonomy also makes data quality non-negotiable
We’ve had automation and self-healing capabilities in networks for years, and deterministic remediation and closed-loop controls have certainly delivered efficiency gains. But traditional automation operates within predefined logic and relatively narrow domains, and what operators are now targeting is autonomous agent-led decisioning. These are systems that can reason across network, service and customer domains and execute bounded decisions within production environments.
But that only works if the underlying data is reliable, and that's where a disciplined data-services practice becomes crucial.
What preparing telco data for AI actually involves
When I refer to using data services to prepare data for autonomous workflows, I’m describing an ongoing, constant engineering discipline – not a one-time clean-up exercise – that typically involves:
- Ingestion and normalization – collecting data from heterogeneous sources, converting protocols, aligning schemas and validating integrity
- Correlation and enrichment – linking network, service and customer domains into a coherent operational context
- Embedded governance – defining ownership, enforcing security policies and maintaining lineage
- Real-time distribution – making contextualized data available to AI agents with predictable latency
Without these capabilities, AI remains an analytical layer sitting on top of fragmented systems. But with them, it can be embedded directly into business and network workflows and dramatically accelerate key outcomes for operators.
Autonomy starts – and often stalls – in the data layer
When I talk to operators about letting AI agents execute business-critical actions, the conversation usually shifts quickly from the model to the plumbing underneath it. In theory, the models are capable, but in practice, the surrounding data architecture often isn't ready, and that's where autonomy gets challenging.
AI is accelerating data creation at an unprecedented rate, but it doesn't automatically resolve the structural weaknesses already present in telecom data estates. In fact, in many cases, it exposes them even faster.
Yes, autonomous agent-led systems are achievable, and the business gains are significant. But they depend on disciplined data services – the engineering work required to prepare and govern the data layer – because AI systems will only ever be as reliable as the data they operate on.
Join us at MWC 2026
Amdocs recently introduced aOS, the world's first agentic operating system purpose-built for telecommunications. At MWC 2026, we’re showcasing how it works in practice.
Agentic Services
The service-as-software platform powering orchestrated multi-agent enterprise workflows
Amdocs aOS
The world’s first agentic operating system purpose-built for telecommunications.
