AI Is Everywhere in Telecom, But Most Pilots Stall

AI is deeply embedded across telecom use cases – from network operations and customer care to marketing and billing automation. Yet despite this surge in experimentation, most GenAI pilots don’t make it past the proof-of-concept stage.

New research from MIT’s Project NANDA reveals that this challenge is not unique to telecom. In fact, 95% of GenAI pilots across industries fail to scale into production. The issue isn’t adoption—it’s execution. Telecom operators are facing the same core barriers: disconnected data, brittle workflows, and tools that don’t learn or evolve with context.

In a highly competitive and cost-sensitive industry, telecom leaders can’t afford to stay stuck in pilot mode. To scale GenAI, operators must move from experiments to platforms—embedding AI into BSS/OSS, NOC/SOC, and network workflows tied to real business outcomes.

Why GenAI Pilots in Telecom Rarely Scale

Most AI initiatives in telecom fail to move beyond innovation labs because they are treated as isolated projects. Use cases are often piecemeal, success metrics are vague, and handover to operations lacks the necessary tooling, runbooks, and budget.

Without ownership in core systems or a roadmap toward operational integration, even promising GenAI pilots are abandoned before scaling.

Executives must shift perspective: AI is not a side initiative—it is a core capability that should be built into the platforms that run revenue, experience, and infrastructure.

The Data Problem Behind AI Underperformance

AI doesn’t work without clean, connected data—and telecom struggles with fragmentation. Data is siloed across RAN, core, transport, CRM, and IT systems. Legacy platforms, incompatible schemas, and event stream gaps make it difficult to train models that can support forecasting, anomaly detection, or ticket deflection.

This is a data engineering challenge, not just a data science one. Without unified telemetry, real-time streaming, and feature stores, even state-of-the-art LLMs will underperform.

Governance, Trust, and Regulatory Pressure

With GenAI expanding into both customer- and network-facing functions, data privacy and compliance risks grow. Tools like chatbots and recommendation engines introduce concerns about hallucinations, bias, and data leakage.

Telecom operators must now meet rising regulatory expectations across regions, particularly under GDPR, CCPA, and new EU AI Act provisions. This makes model governance, auditability, and risk management non-negotiable for deployment.

Failing to address compliance requirements early is a leading reason why telecom GenAI pilots are shelved before launch.

From Pilots to Platforms: Embedding AI in Telecom Systems

To scale AI, telecom leaders are embedding intelligence into the systems that run the business—not just testing AI in isolation.

Embedding AI into BSS/OSS Workflows

Vendors like Cerillion are evolving their BSS/OSS offerings with a “bring your own AI” model, enabling integration of both public and private LLMs. Subex reports significant reductions in billing inquiries using AI-powered automation. Embedded AI shortens time to value, simplifies integration, and increases operational trust.

Using GenAI Across Product, Marketing, and Customer Support

GenAI is compressing time-to-market. Cerillion’s image-to-configuration tool can turn whiteboard sketches into product catalog entries in minutes. AI is also helping marketing teams refine campaign targeting and optimize promotions.

In customer service, assistants like Vodafone‘s TOBi now handle a majority of incoming queries in some markets—improving CSAT, reducing agent load, and boosting resolution rates.

Edge AI for Network Automation

On the infrastructure side, pairing AI with edge computing enables real-time use cases such as closed-loop optimization, self-healing, and dynamic resource allocation. As 5G Standalone, IoT, and network slicing expand, Edge AI will become foundational.

SK Telecom, for example, has restructured around AI platforms and agent-based automation—addressing tasks from botnet detection to traffic prediction.

Market Outlook: Telecom AI Through 2033

Analyst forecasts vary in dollar value, but the trend is clear—telecom AI is heading toward multi-billion-dollar scale. Growth areas include:

• Network automation and assurance
• Predictive analytics
• Generative AI for marketing and operations
• Real-time decision-making at the edge

Regions like North America and Asia-Pacific are leading in adoption due to their cloud readiness and infrastructure investment capacity. However, European telcos face profitability constraints that limit AI scaling—even as operational efficiency becomes more urgent.

Shifting the AI Operating Model

To overcome organizational inertia, telecom leaders are evolving their operating models. Some operators, like SK Telecom, are spinning out dedicated AI business units with separate P&Ls and product roadmaps.

This “company-in-company” model allows for faster delivery, consolidated talent, and AI-native platform thinking—breaking free from legacy IT constraints.

How to Scale Telecom GenAI: A Pragmatic Blueprint

1. Build a Unified Data Plane

To unlock the full value of GenAI, telecom operators must begin with data. A unified, AI-ready data architecture enables consistent model training, operational reliability, and scalable automation across the business.

• Align telemetry across RAN, transport, core, CRM, and IT.
• Invest in real-time streaming, common data schemas, and feature stores.
• Apply policy, quality, and lineage controls to support GenAI workloads.
• Use TM Forum Open APIs and Open Digital Architecture to reduce integration friction.

2. Modernize BSS/OSS and Interfaces

Outdated BSS/OSS systems are one of the biggest blockers to GenAI scale. Modern, cloud-native, and API-first platforms allow for better orchestration, faster deployments, and improved AI lifecycle management.

• Shift to cloud-native, API-first platforms.
• Adopt containerized microservices and GitOps workflows.
• Implement MLOps and ModelOps for testing and rollback.
• Enable inference at the edge for latency-sensitive tasks in assurance and security.

3. Govern for Safety and Compliance

As GenAI becomes deeply embedded in critical workflows, telcos must proactively address data privacy, regulatory compliance, and AI safety. Governance frameworks are key to de-risking deployments.

• Build model risk management frameworks.
• Include explainability, bias testing, and human-in-the-loop checkpoints.
• Enforce privacy-by-design principles across data and model pipelines.
• Test SOC/NOC automations under adversarial and red-team scenarios.

4. Link AI to ROI and P&L Outcomes

GenAI only scales when it delivers measurable value. By tying AI initiatives directly to operational KPIs and financial performance, telecom leaders can move from experimentation to business impact.

• Focus on measurable impact: reduced MTTR, lower churn, fewer disputes, higher uptake.
• Set KPIs and baselines before deployment.
• Fund cross-functional teams responsible for benefit realization, not just delivery.

5. Close the AI Skills Gap in Telecom

Successful GenAI deployments depend on talent. Telecom operators must build in-house capabilities in AI engineering, network science, and automation—or risk falling behind digitally mature competitors.

• Upskill teams in AI engineering, network data science, and SRE/NRE.
• Partner with hyperscalers, ISVs, and integrators.
• Engage in TM Forum Catalysts and joint R&D to accelerate learning.

How to Scale Telecom GenAI: A Pragmatic Blueprint

1. Build a Unified Data Plane

• Align telemetry across RAN, transport, core, CRM, and IT.
• Invest in real-time streaming, common data schemas, and feature stores.
• Apply policy, quality, and lineage controls to support GenAI workloads.
• Use TM Forum Open APIs and Open Digital Architecture to reduce integration friction.

2. Modernize BSS/OSS and Interfaces

• Shift to cloud-native, API-first platforms.
• Adopt containerized microservices and GitOps workflows.
• Implement MLOps and ModelOps for testing and rollback.
• Enable inference at the edge for latency-sensitive tasks in assurance and security.

3. Govern for Safety and Compliance

• Build model risk management frameworks.
• Include explainability, bias testing, and human-in-the-loop checkpoints.
• Enforce privacy-by-design principles across data and model pipelines.
• Test SOC/NOC automations under adversarial and red-team scenarios.

4. Link AI to ROI and P&L Outcomes

• Focus on measurable impact: reduced MTTR, lower churn, fewer disputes, higher uptake.
• Set KPIs and baselines before deployment.
• Fund cross-functional teams responsible for benefit realization, not just delivery.

5. Close the AI Skills Gap in Telecom

• Upskill teams in AI engineering, network data science, and SRE/NRE.
• Partner with hyperscalers, ISVs, and integrators.
• Engage in TM Forum Catalysts and joint R&D to accelerate learning.

2025 AI Priorities for Telecom Executives

Move From Pilot to Platform

The GenAI Divide—where pilots thrive but production fails—is closing in fast. The window for experimentation is narrowing. As enterprises across industries begin locking in GenAI partners, telecom operators must act now.

Executive Actions to Take

• Treat AI as an operational principle, not a tech demo.
• Fund the data plane as strategic infrastructure.
• Embed AI into core workflows, not isolated interfaces.
• Staff for platform execution, not just experimentation.

The operators who industrialize GenAI today will benefit from cost savings, faster time-to-market, and improved customer experience—while those stuck in pilot mode risk falling behind.