Verizon Connect scales agentic AI to analyze 500M daily data points for 100K users

Verizon Connect, a global fleet management provider, faced a massive data overload challenge. With over 1.2 million active vehicle subscriptions generating more than 500 million data points daily across 80,000 unique indicators, fleet managers were drowning in fragmented logs and reactive spreadsheets. To transform this chaos into actionable insights, they chose an agentic AI approach rather than static dashboards or rule-based systems. Their solution combines a serverless statistical anomaly detection module (built with AWS Step Functions and Lambda) that handles computational heavy lifting, followed by AI agents that use the open-source Strands Agents SDK to investigate anomalies, ask follow-up questions, and synthesize coherent narratives. This architecture runs in a serverless environment, scaling horizontally to process different customers or data segments in parallel.

The system first ingests raw data and identifies specific anomalies using specialized statistical code, avoiding the accuracy pitfalls of asking LLMs to parse large tabular datasets. Then, AI agents reference both the anomaly output and raw data to reason about why the anomaly occurred and what actions to recommend. The final insights are delivered to users through the Reveal application. The result: 100,000 daily users now receive clear, actionable alerts about emerging safety issues, maintenance needs, and operational inefficiencies. By offloading numerical analysis to dedicated models and letting agents focus on contextual reasoning, Verizon Connect achieved scalability, cost-efficiency, and a dramatic reduction in manual data hunting.