Session

How Databricks and TM Forum are Dialing up the AI-Native Telco

Overview

Experience	In Person
Track	Analytics & BI
Industry	Communications, Media & Entertainment
Technologies	Genie
Skill Level	Beginner
DOWNLOAD SESSION SLIDES

In an industry where network reliability and customer experience are sacred, telecommunications leaders need accurate, real-time insights fast. Yet business and data teams have long existed in silos, keeping these crucial insights out of reach for those who need it in-the-moment the most. But now, operators are able to break down these silos with AI-powered conversational business intelligence, made possible by solutions like AI/BI Genie.

In this session, discover how Genie is empowering business users across telecommunications to get the answers and insights they need from their data in natural language, allowing them to ask complex questions and receive instant, governed insights. Learn how experts at leading operators including Frontier, GCI, and DIRECTV conversational intelligence are using Genie to eliminate traditional data bottlenecks and identify new strategies to move their business forward. Attendees will walk away with a blueprint for scaling AI-driven conversational BI and get a firsthand look at how Genie is enabling real telecom organizations to do more with their data.

Session Speakers

IMAGE COMING SOON

Andy Flint

/RVP Communications
Databricks

Felix Perez

/Executive Director Data Analytics and AI
SBA Communications

IMAGE COMING SOON

Guy Lupo

/EVP, Trustworthy AI & Data
TM Forum

Ilya Shestakov

/Director, Data Science & Analytics
GCI

Jonathan Sunberg

/Sales Director - Telco
Databricks

Layla Yang

/Principal Solution Architect
Databricks

Full Summary

Inside TM Forum and Databricks: Building agentic telecom with open standards and governed AI

Telecom operators manage sprawling data across orders, networks, billing, and customer systems. The conversation outlines how TM Forum's open standards and code-first collaboration model combine with the Databricks Platform to turn that data into action through governed, domain-specific agents. Demonstrations on the Databricks Platform for communications show agents compressing analyst work into moments, surfacing cross-domain insights, and safely closing the loop on network operations. Operators exploring how AI reshapes their analytics stack will find relevant context in the complete guide to business intelligence and analytics in the AI era. The payoff is faster reasoning across silos, traceable decisions, and safer autonomy. Operators can move toward TM Forum's Autonomous Networks Level 4 while keeping humans responsibly in the loop.

TM Forum's code-first catalysts and the Canvas operator model

TM Forum, the connectivity industry's alliance, advances a large Open Digital Architecture used by carriers worldwide, including frameworks such as eTOM and widely adopted open APIs. Its Catalyst program pairs proposing members with multiple operators to co-build proof-of-concept implementations that are open, reusable, and grounded in code, not slideware.

The Trustworthy AI and Data mission pushes members to "vote with code." TM Forum's Canvas, a Kubernetes environment with standardized APIs, now supports operators that make it easier to integrate agents with billing, inventory, and other domains. Databricks contributed a data operator to Canvas so agents can work against governed, standardized data services, accelerating reuse across projects.

Domain-specific Genie agents that reason across telco data

Genie sits on top of an open lake, is governed by Unity Catalog, and connects through the Genie Ontology so agents understand both data and meaning. Rather than one monolithic assistant, teams create domain Genies for customer care, network, and billing, then coordinate them with a supervisor agent for cross-domain questions.

A CFO view illustrated the shift from dashboards to decisions. The agent proactively highlighted a widening loss in business wireline, connected it to legacy services decline, and suggested fiber bundling actions. Order-to-activate delays were traced to manual B2B provisioning, a surge from a new device launch, and a billing cutover that stressed legacy systems. Postpaid churn was tied to installment plan expiries with a recommendation for proactive retention triggers. The agent also modeled how many net-new fiber subscribers would be required annually to offset legacy voice decline. Every answer was grounded in the lakehouse data and business logic, turning passive KPIs into trusted guidance.

From detection to safe action in the network

A live, simulated radio access network within Databricks streamed telemetry that, when degraded, triggered an agent-led triage. The agent classified an RF path issue, mapped affected sites and customers, quantified revenue impact, and retrieved relevant runbooks and vendor manuals using AT&T's open-source OTEL models. Before taking action, it simulated remediation against a digital twin to confirm neighboring sites had capacity, then issued a coverage compensation order and, when needed, dispatched the right technician based on location, inventory, and skills.

Safety and governance anchored the design. The LLM never touched the network directly, selecting tools that propose changes to the service management orchestration layer. Faster models handled triage while telecom-tuned open models corroborated responses, a model-as-a-service pattern that reduces sprawl. Every tool call, retrieval, citation, and safety judgment was captured as an MLflow trace, data was governed in Unity Catalog, and activity was queryable through Genie for full observability.

Standards plus governance make agentic telecom production-ready

Open models and APIs supply a shared language, domain-specific agents deliver relevant reasoning, and end-to-end observability builds trust. That combination moves operators from one-off demos to repeatable, auditable patterns that support TM Forum's vision for autonomous networks, with humans supervising responsibly and agents executing detection, diagnosis, and safe remediation.

From dashboards to decisions: how telcos operationalize AI safely

Operators that anchor agents in shared standards, govern data and models centrally, and insist on traceability can turn intent into safe automation. The result is faster cross-domain insight, fewer swivel-chair workflows, and a credible path to autonomy without sacrificing control.

FAQ

Catalysts are code-based proofs of concept co-built by a proposing member and multiple operators under TM Forum's bylaws. Projects are showcased at TM Forum events, often win awards, and are made available to members and registered non-members, with the option to extend or experiment in the TM Forum innovation hub.

Genie Ontology captures the meaning and relationships behind data, not just tables and columns. That context lets domain agents ground responses in both the underlying lake data and the business logic, enabling reliable cross-domain reasoning for customer, network, and billing scenarios.

Models are provided centrally as a governed service rather than spun up ad hoc in each business unit. In the network demo, faster general-purpose models handled triage, while AT&T's open-source telecom OTEL models corroborated responses, all under a single governed layer.

The LLM does not directly control the network. It selects tools that propose intent to the service management orchestration layer, simulates proposed changes against a digital twin, verifies neighboring capacity, and only then issues policy guidance or dispatches technicians.

The team highlighted contributions to multiple Catalysts, including InsightX and model-as-a-service work, to be shown at TM Forum's flagship event in Copenhagen. Attendees were invited to the industry solution booth for deeper walkthroughs.