FinOps-Driven Cloud Cost Optimization in Electric Vehicle Telematics Platforms: A Longitudinal Case Study with Machine Learning Forecasting

Electric vehicles (EVs) are progressively reliant on cloud services for data storage, connectivity, and real-time analytics. However, despite extensive research on EV cloud architectures, a persistent gap remains between conceptual frameworks and applied, repeatable strategies for controlling cloud operational costs in production environments. This study addresses that gap by applying a FinOps (Cloud Financial Operations) methodology to an EV telematics platform, demonstrating how systematic cost optimization can be achieved without compromising performance. The FinOps approach, which encompasses the Inform, Optimize, and Operate phases, was implemented with three objectives in mind: The primary objectives are threefold: firstly, to gain visibility into cloud usage and cost drivers; secondly, to deploy cost-saving technical measures; and thirdly, to establish continuous governance. A 40-month case study was conducted with the objective of reconfiguring the platform's cloud deployment and quantifying savings against counterfactual baselines. A forecasting framework combining multiple time series approaches (polynomial regression, ARIMA, Facebook Prophet, and LSTM neural networks) was utilized to project cost trajectories and support attribution of savings. The polynomial model relating cloud cost to vehicle count achieved an R² of 0.996 with a mean absolute percentage error (MAPE) of 2.05%, while the Prophet model achieved an in-sample MAPE of 8.37%. Key interventions included software optimization (migration to .NET 6), rightsizing and reserving cloud resources, and database query tuning guided by Request Unit (RU) metrics. These interventions resulted in an 88% reduction in cost per vehicle and a 55% decrease in total monthly cloud expenditure, despite the growth of the connected EV fleet by over 340%. In EV telematics, a “message” is a single telemetry event sent by a vehicle and processed in the cloud; at fleet scale, volumes can quickly reach billions even with modest reporting rates (e.g., 10,000 vehicles every 5 s → ~172.8M/day, >5B/month). Therefore, we normalize spending as cost per billion messages, which decreased by 95.5% (a 22.3× efficiency gain). The findings demonstrate that FinOps can materially reduce cloud operating costs in production connected-vehicle platforms. While cloud spend may represent a small fraction of total vehicle TCO, the proposed approach offers a replicable framework that links FinOps practices to measurable, real-world outcomes and supports scalable cost governance as platforms grow.