Sustainable and ROI-Driven Predictive Analytics: Measuring Economic and Environmental Value in Enterprise Deployments

Abstract Enterprises deploying predictive analytics at scale face a dual mandate: maximizing financial returns while constraining Scope 2 and Scope 3 carbon emissions. This study presents a longitudinal, mixed-methods analysis spanning January 2024 to September 2026, examining carbon-aware scheduling and blended return-on-investment (ROI) frameworks across finance, manufacturing, and retail sectors. Using anonymized deployment data from 18 enterprise environments, CodeCarbon energy telemetry, and a novel 'Model P&L' attribution methodology, we quantify the simultaneous economic and environmental impacts of transitioning from standard inference pipelines to carbon-aware dynamic routing architectures. Key findings indicate a 278% mean blended ROI improvement, a 38.1% reduction in inference-associated carbon emissions, and a 4.9 percentage-point accuracy gain across sectors by Q3 2026. Carbon cost internalization reduces net model P&L drag by up to 47%. These results provide actionable frameworks for Chief Information Officers (CIOs), sustainability officers, and AI governance teams navigating the accountable AI imperative in 2026. Keywords: sustainable AI, carbon-aware scheduling, blended ROI, model P&L, predictive analytics, green computing, Scope 3 emissions, accountable AI, enterprise AI governance, longitudinal analysis