Data Lakes Are Not Enough: Why AN-4 Autonomous Networks Need an Operational Intelligence Layer



For years, telecom operators have invested heavily in data platforms. Data lakes became the destination for virtually every operational dataset: telemetry, alarms, inventory, performance metrics, customer experience data, and configuration changes. The assumption was straightforward: consolidate the data first, then apply analytics and AI to extract value. That strategy has delivered real benefits. Reporting is better. Analytics are more comprehensive. Machine learning has become easier to scale. But there’s a growing disconnect between what these platforms were designed to do and what autonomous networks now require.


As the industry works toward TM Forum's AN-4 vision, the objective is no longer to provide engineers with better information, it’s to enable software to make operational decisions that can be trusted. That’s a very different problem. The obstacle isn’t a lack of data - most operators already have more operational data than they can reasonably consume - the challenge is creating a trustworthy representation of the network as it exists now, not as it existed five minutes ago or according to five different OSS systems.


Data Lakes Store Data. Networks Operate on State

A modern network is constantly changing. Configuration updates are pushed every minute. Services are activated and modified. Traffic patterns shift. Devices fail and recover. Customers move between access technologies. Software versions change. Topology evolves. Every one of these events generates data, and data lakes are very good at preserving it.

What they don’t provide is a continuously validated understanding of the current operational state of the network. That distinction matters. An autonomous system deciding whether to reroute traffic, isolate a fault, or initiate a repair can’t work from a collection of loosely related historical records. It needs confidence that the information it’s using reflects reality at the moment the decision is made.


The Questions AI Must Answer

At AN-4, AI is expected to move beyond recommending actions; it must execute them. But before doing so, it has to answer a series of questions that operators have traditionally answered themselves:

  • What’s happening right now?
  • Which customers and services are actually affected?
  • What’s the underlying cause?
  • What’s the safest corrective action?
  • Did the action solve the problem without creating another one?

These aren’t analytical questions, they’re operational questions. Answering them requires far more than access to historical information. It requires context, relationships, timing, and continuous validation.


Where Data Lakes Reach Their Limits

Data lakes remain essential infrastructure, but they were built for enterprise analytics rather than operational decision-making. They prioritize persistence over current state. They consolidate data but not agreement between systems. They correlate events but rarely capture causality. And they execute workflows without continuously verifying outcomes.


The conclusion should not be that operators need to replace their data lakes. The industry has invested too much in enterprise data platforms to abandon them. They remain indispensable for analytics, regulatory reporting, capacity planning, AI model development, and long-term business intelligence. What’s missing is a dedicated operational intelligence layer.


Positioned above the data lake, this layer continuously reconciles information from operational systems, maintains an accurate Digital Twin of the network, models dependencies between infrastructure, services, and customers, and validates the outcome of automated actions as they occur. Its purpose isn’t to store more information, it’s to establish operational truth.


From Information to Decisions

The path to AN-4 is often presented as an AI challenge. In practice, it’s largely an architectural one. 

The telecom industry has spent the last twenty years building platforms that help engineers answer questions. The next phase will focus on platforms that enable software to make decisions safely.


Historical repositories remain essential for understanding the past. Operational intelligence provides something different: a trusted, continuously updated view of the present. Operators that succeed in autonomous networking won’t necessarily have the most sophisticated AI models, they’ll be the ones that provide those models with an accurate, reconciled, continuously verified understanding of the network they are expected to manage.


Data lakes will remain an important part of that architecture, but they’re simply no longer sufficient on their own. The next generation of autonomous networks will be built not on larger repositories of operational data, but on systems capable of transforming that data into trusted operational intelligence. That is the foundation AN-4 ultimately demands.


Coming next: Why NOCs Need AI Teammates, Not AI Assistants. Most AI tools wait for an engineer to ask the right question. Autonomous networks can’t afford to wait. The next generation of Network Operations Centers will rely on AI that continuously observes, reasons, collaborates, and acts alongside engineers, transforming operations from reactive troubleshooting to proactive network management.