Vodafone implemented a comprehensive AI and GenAI strategy to transform their network operations, focusing on improving customer experience through better network management. They migrated from legacy OSS systems to a cloud-based infrastructure on Google Cloud Platform, integrating over 2 petabytes of network data with commercial and IT data. The initiative includes AI-powered network investment planning, automated incident management, and device analytics, resulting in significant operational efficiency improvements and a planned 50% reduction in OSS tools.
Vodafone, one of the world’s largest telecommunications companies, is undertaking a comprehensive transformation of its network operations using AI and GenAI technologies in partnership with Google Cloud. This case study, presented as a conversation between Simon Norton (Head of Digital Network and OSS practice at Vodafone) and Brian Kracik from Google Cloud, outlines how Vodafone is applying generative AI to enhance customer experience through improved network management, operations, and investment planning.
The conversation reveals that Vodafone’s AI journey in the network domain began approximately four to five years ago with traditional AIOps implementations, and they are now evolving to incorporate GenAI capabilities across the entire network lifecycle. It’s worth noting that this is a promotional interview hosted by Google Cloud, so the perspectives shared are inherently positive about the partnership and technology choices.
Vodafone faced several interconnected challenges that necessitated a transformation of their network operations approach:
The first major challenge was dealing with legacy OSS (Operational Support Systems). Like most established telecoms, Vodafone had accumulated a substantial base of legacy tools that made it difficult to extract insights and correlate information across different systems. Simon Norton described the challenge of trying to understand the “insight action cycle” when data was fragmented across IT systems, commercial systems, and network systems.
The second challenge involved understanding customer experience as a multivariate problem. Norton emphasized that great or poor customer experience isn’t driven by a singular event but rather a “linear chain of events with many variables.” The network is a huge variable in this equation, but pinpointing exactly where network interventions should be made to improve experience was historically difficult.
A third challenge was the inability to correlate data across previously siloed on-premises deployments of performance management tools, device analytics, and customer experience management platforms. Norton explicitly stated that building applications on top of correlated data “simply wasn’t possible” with their previous infrastructure.
Vodafone’s approach to solving these challenges involves multiple layers of technical implementation, with GenAI serving as a key capability on top of a solid data foundation.
The foundational decision that Norton describes as “prescient” was Vodafone’s strategic partnership with Google Cloud, which began approximately five years ago. Today, they have:
This data platform serves as the foundation upon which both traditional AI/ML and GenAI applications are built. Without this unified data layer, the GenAI applications would lack the context and information needed to be useful.
One specific product mentioned is “Net Perform,” which was migrated to GCP in December. This advanced device analytics capability enables Vodafone to understand exactly how customers experience the network from their devices. The platform allows correlation of device-level data with traditional network feeds from customer experience management platforms and performance management platforms. This enables more personalized customer experiences and powers capabilities like guided diagnostics journeys for customers.
The most significant LLMOps application discussed is the use of GenAI as a copilot for complex incident management. Norton described this as follows: when closed-loop automation using traditional AI/ML methods cannot handle an incident, GenAI becomes “brilliant to copilot the human” by:
The value proposition is straightforward: “Less problems, fixed more quickly? Better customer experience.”
This represents a classic copilot pattern in LLMOps where the LLM augments human decision-making rather than fully automating it. This is particularly appropriate for complex incident management where edge cases are common and the consequences of errors are significant.
Another GenAI application under development is what Vodafone calls “smart CapEx” - a GenAI-powered next-generation network investment planning capability. This requires integration of:
Norton emphasized that building such a product requires fundamentally different ways of working, bringing together data sets and expertise from multiple teams. His team may not be expert on commercial data sets, while commercial teams may not understand network data - the GenAI application requires both.
A major infrastructure modernization effort involves creating “Unified Performance Management” - a single system to replace over 100 traditional performance management systems across different network domains. This consolidation:
The broader context for these GenAI initiatives is a massive OSS simplification and modernization program. In the next three years, Vodafone plans to:
Norton acknowledged that while GenAI is exciting, the underlying OSS is “mission critical in its nature” and if legacy persists, it will hinder the ability to achieve higher-level AI objectives.
An interesting aspect of this case study is the emphasis on organizational change rather than just technology. Norton repeatedly stressed the importance of cross-functional collaboration:
This reflects a common theme in LLMOps implementations: the technology is often the easier part, while organizational alignment and cultural change present greater challenges.
While this case study provides valuable insights into how a major telecom is approaching GenAI in network operations, it’s important to note several limitations:
The interview is promotional in nature, hosted by Google Cloud and featuring their customer. Specific quantitative results or metrics are largely absent. We hear about plans and intentions (e.g., the 600 tool reduction target) but limited evidence of achieved outcomes from the GenAI implementations specifically.
The GenAI applications described - particularly the incident management copilot - are presented as aspirational use cases rather than fully deployed production systems. Norton uses phrases like “I really think that AI operations, coupled with GenAI, is going to be game changing for us” which suggests these capabilities are still emerging.
The technical details of how GenAI models are trained, deployed, monitored, and maintained in production are not discussed. Questions about model selection, fine-tuning, prompt engineering, evaluation metrics, and governance are not addressed.
That said, the case study does illustrate a pragmatic approach to LLMOps: building on a solid data foundation, starting with traditional AI/ML and adding GenAI where it provides unique value (particularly in human-in-the-loop scenarios), and recognizing that organizational change is as important as technical implementation. The emphasis on using GenAI for copiloting complex decisions rather than full automation reflects a mature understanding of where current LLM capabilities are most effectively applied.
Snorkel developed a specialized benchmark dataset for evaluating AI agents in insurance underwriting, leveraging their expert network of Chartered Property and Casualty Underwriters (CPCUs). The benchmark simulates an AI copilot that assists junior underwriters by reasoning over proprietary knowledge, using multiple tools including databases and underwriting guidelines, and engaging in multi-turn conversations. The evaluation revealed significant performance variations across frontier models (single digits to ~80% accuracy), with notable error modes including tool use failures (36% of conversations) and hallucinations from pretrained domain knowledge, particularly from OpenAI models which hallucinated non-existent insurance products 15-45% of the time.
Ericsson's System Comprehension Lab is exploring the integration of symbolic reasoning capabilities into telecom-oriented large language models to address critical limitations in current LLM architectures for telecommunications infrastructure management. The problem centers on LLMs' inability to provide deterministic, explainable reasoning required for telecom network optimization, security, and anomaly detection—domains where hallucinations, lack of logical consistency, and black-box behavior are unacceptable. The proposed solution involves hybrid neural-symbolic AI architectures that combine the pattern recognition strengths of transformer-based LLMs with rule-based reasoning engines, connected through techniques like symbolic chain-of-thought prompting, program-aided reasoning, and external solver integration. This approach aims to enable AI-native wireless systems for 6G infrastructure that can perform cross-layer optimization, real-time decision-making, and intent-driven network management while maintaining the explainability and logical rigor demanded by production telecom environments.
Predibase, a fine-tuning and model serving platform, announced its acquisition by Rubrik, a data security and governance company, with the goal of combining Predibase's generative AI capabilities with Rubrik's secure data infrastructure. The integration aims to address the critical challenge that over 50% of AI pilots never reach production due to issues with security, model quality, latency, and cost. By combining Predibase's post-training and inference capabilities with Rubrik's data security posture management, the merged platform seeks to provide an end-to-end solution that enables enterprises to deploy generative AI applications securely and efficiently at scale.
