From data overload to actionable insights: How Verizon Connect scaled agentic AI to 100,000 users

Background

Fleet managers at Verizon Connect, a global provider serving businesses via its Reveal platform, faced a core operational crisis: data overload. Managing thousands of vehicles each generated hundreds of daily data points, resulting in over 500 million data points across 80,000 unique indicators daily. This scale made manual analysis—relying on fragmented paper logs and reactive spreadsheets—impossible. The inability to identify emerging safety issues, maintenance needs, or operational inefficiencies proactively led to costly problems. The challenge was not a lack of data but the inability to extract timely, actionable intelligence from it.

Key Points

1. Strategic Choice of Agentic AI Over Static Tools
Verizon Connect rejected building a static dashboard or rule-based automation system. Such tools only catch predefined patterns and cannot adapt to the unpredictable, complex nature of fleet operations. Instead, they chose agentic AI because it can dynamically investigate new patterns, ask follow-up questions, and adapt its analysis in real-time. This represents a shift from monitoring for known issues to actively discovering unknown ones.

2. Scalable, Hybrid Architecture for Cost-Efficiency
The solution's architecture is designed for scale and cost-efficiency, separating computationally intensive tasks from AI reasoning.

• Anomaly Detection Module: This is the critical first layer. A common pitfall is asking an LLM to perform numerical analysis on raw tabular data, which it struggles with at scale. Verizon Connect built a serverless statistical model (using AWS Step Functions and Lambda) to handle this heavy lifting. It processes structured data to identify specific anomalies and writes them to a dedicated table.
• AI Agent Activation: Once anomalies are ready, the system triggers multiple AI agents in parallel, each assigned to a different customer or data segment. This parallel processing improves performance and scalability.
• Reasoning & Context Layer: This is where the agentic intelligence operates. An AI agent orchestrates the final analysis by querying the anomalies table for the "what" (the identified problem) and cross-referencing the raw data store for the "why" (the underlying cause). It then uses an LLM to synthesize these inputs into a coherent, human-readable narrative.

3. From Data Overload to Delivered Insight
The final output is a generated insight stored and served to the end-user via the Reveal application. The process transforms abstract data points into concrete, actionable information for fleet managers. The daily trigger initiates this entire workflow, ensuring insights are current and proactive.

Significance

Verizon Connect's implementation demonstrates a practical blueprint for solving the "data-to-insights" transformation at enterprise scale.

• Addresses the LLM Limitation Directly: By offloading numerical analysis to specialized code, the design avoids LLM weaknesses with raw data tables, ensuring accuracy and efficiency. The LLM is then used for its core strength: language-based reasoning and synthesis.
• Enables Proactive Operations: The system shifts fleet management from a reactive model (responding to problems) to a proactive and predictive one. Identifying anomalies before they escalate allows for preventive maintenance, enhanced safety, and operational optimization.
• Provides a Replicable Model: The architecture—decoupling data processing from AI reasoning, using parallel agents, and focusing the LLM on contextual interpretation—is a transferable pattern for other industries drowning in complex operational data. It showcases how to build a scalable, cost-effective system that turns data from a burden into a strategic asset.