The Challenge: Data Without Context in the Telco AI Era

GenAI and analytics tools and solutions pull data from applications and from downstream and external data sources. However, when data is decontextualized from the applications where it was created, it often loses the business meaning, relationships, and rules that were embedded in the original application. This information is critical for producing dependable, business-relevant output using AI. For example, in the sales domain, customers might be called an account, a consumer, in the network domain will be mostly called subscriber. Different business units might use a different denomination. Line of business systems might have their own naming conventions; external entities might follow telecom-specific standards.

Telecom operators are increasingly constrained not by lack of data, but by lack of data meaning:

• Network data (RAN, core, transport).
• IT/OSS/BSS data.
• Customer interaction data (apps, call centers, digital channels).
• External ecosystem data (partners, devices, edge).

While these datasets are abundant, they are:

• Fragmented across systems.
• Largely unstructured, with proprietary formats or inconsistently modeled.
• Needs improvement in terms of governance in terms of quality and semantics.

As a result:

• Humans find challenging to decipher the meaning, or to extract insights - for example: what are the numbers of consumers being affected by a drop of the QoS KPI p95_latency_ms that affects apps latency in real time.
• AI models struggle to interpret relationships across domains.
• Use cases (e.g., autonomous networks, customer 360, fraud detection) require heavy manual data engineering.
• Scaling GenAI and agent-based workflows becomes complex.

This reflects a broader issue: when data is extracted from its originating system, it loses its business meaning, relationships, and operational rules.

This semantic consistency creates a shared language across the enterprise and ecosystem, independent of specific formats or proprietary structures. The ontology manages translation, meaning, and relationships, keeping business logic simple and portable. As a result, it provides the foundation for AI to reason, plan, and execute consistently. The ontology (preview) item (part of the Fabric IQ (preview) workload) digitally represents the enterprise vocabulary and semantic layer that unifies meaning across domains and OneLake sources[1].

 

“Ontology modeling recognizes business concepts over table schemas[2].”

 

Why Telcos need an Ontology

Some organizations are investing in a robust ontological layer to address this problem. A semantic layer is a system of technologies and techniques that organizes and maintains a consistent and unified representation of data from various sources that is interpretable by humans and machines. Functioning conceptually between data and the user (human or machine), a semantic layer describes data’s business meaning, its relationships with other data, and the rules the data must adhere to.

 

“Microsoft Fabric IQ provides a way to define business vocabulary in an ontology and bind the ontology to data sources[3].”

 

An ontology layer in telecom functions as a “digital understanding layer” for enterprise data:

• Describes network entities (tower cells, sites, devices, slices).
• Defines customer concepts (subscriber, household, product bundles).
• Encodes relationships (e.g., customer → service → network resource).
• Helps to enforce business and regulatory rules (e.g., data privacy, SLA constraints).

This layer enables:

• Cross-domain reasoning (network ↔ customer ↔ service).
• More consistent AI outputs.
• Faster onboarding of new data sources and use cases.

 

Figure 1 - An ontology (preview) related to mobile networks

 

Critically, in the GenAI era, AI requires machine-readable context (e.g., ontologies, knowledge graphs) to generate reliable outputs at scale.

 

“In Microsoft Fabric IQ, ontology modeling differs from traditional analytical modeling by starting with business concepts rather than specific use cases[4].”

 

Your ontology is a graph database—a data structure built from two fundamental elements: nodes and edges. In the graph view below, each colored circle is a node (a real record—a base station, a network, and each labeled arrow connecting them is an edge (a relationship).

 

Figure 2 - An ontology (preview) represented as a graph that can be queried by a data agent

 

The target state for an ontology layer for Telecom AI

GenAI raises the bar for accessing contextual metadata at speed and scale. Without that context, GenAI tools are less able to produce relevant, reliable responses for business users and are more likely to misinterpret data.

To address this, organizations need modern capabilities such as semantic search, which goes beyond keyword matching by using metadata to understand the intent behind a query. They also benefit from knowledge graphs, which provide a machine-readable representation of enterprise knowledge for applications such as network anomaly agents, recommendation engines, fraud detection, and customer experience agents.

An ontology layer captures an organization’s understanding of data, regardless of where it resides or how it is structured. In the GenAI era, it must hold the context, including data usage constraints, permitted combinations, personal data handling rules, and compliance requirements. It can also encode expertise such as data quality judgments and approved exceptions to business rules. With this semantic layer, organizations can scale AI without redesigning governance for each use case, improving the accuracy of AI agents and GenAI models while reducing risk.

 

Microsoft Fabric IQ offers the following capabilities[5]:

• Ontology items define your business vocabulary and bind concepts to data sources.
• Data agents answer natural language questions across multiple data sources.
• Graph stores and queries connected data using relationships.
• Semantic models provide a starting point for generating ontologies from existing Power BI models.

 

 

 

Figure 3 - A data agent querying data that uses a Microsoft Fabric IQ ontology as its data source

 

A modern telecom ontology should:

• Represent enterprise knowledge.
  • Describe the elements that form the network topology.
  • Service portfolio hierarchies.
  • Customer relationships.
  • Unified metadata across OSS, BSS, and IT domains.
• Enable AI-Driven Data Access
  • Semantic search over telecom data (intent-based vs keyword-based).
  • Context-aware for GenAI (prb_util_pct above 80, means RAN congestion and a troubleshoot agent needs to be enacted).
  • Agent interoperability via shared ontology.
• Encoding Governance and Compliance
  • Data usage policies (e.g., GDPR, NIS2, lawful intercept constraints).
  • Permitted data combinations.
  • Data access rules for AI agents.
• Capture Telco Domain Expertise
  • Codified engineering rules (e.g., network thresholds, fault correlations).
  • Data quality scoring and trust levels.

A well-designed ontology layer transforms telecom data from fragmented technical signals into actionable, AI-ready knowledge assets. Microsoft’s strategic partners, such as Kenmei, Infosys and Tech Mahindra, are already innovating and bringing in the market ontology-driven solutions for the telecom industry.

Conclusion

For telecom operators, investing in a robust semantic layer means building data assets that can be understood and trusted at speed and scale. In the GenAI era, this machine-readable representation of enterprise knowledge is important for enabling AI to interpret data, reason effectively, and create operational advantage.

Ontology offers major transformative potential for telecoms, with the ability to reshape operations in ways few other technologies can. While it may seem abstract and at times misaligned with initiatives such as data estate modernization, it is a foundational capability for operators to move toward AI-native operations. The semantic consistency, contextual intelligence, and action-oriented structure provides that it is difficult to be replicated by traditional data management approaches.

 

[1] https://learn.microsoft.com/en-us/fabric/iq/ontology/overview

[2] https://learn.microsoft.com/en-us/fabric/iq/ontology/overview

[3] https://learn.microsoft.com/en-us/training/modules/understand-fabric-iq-fundamentals/

[4] https://learn.microsoft.com/en-us/training/modules/understand-fabric-iq-fundamentals/

[5] https://learn.microsoft.com/en-us/training/modules/understand-fabric-iq-fundamentals/