Carbon-Aware Sustainable Digital Shopping: A Federated, Behavior-Aware System for Real-Time Basket-Level Emissions Optimization in E-Commerce

Consumer food and retail purchases account for a large share of household greenhouse gas emissions. Despite growing interest in sustainability, most e-commerce platforms do not provide real-time carbon information at checkout, when consumers make their final purchasing decisions. This paper presents Carbon-Aware Checkout (CAC), a system that combines life-cycle assessment, machine learning, uncertainty modeling, optimization, and large language models through the Model Context Protocol (MCP). CAC delivers real-time basket-level carbon scores, behavior-adjusted emissions forecasts, cost-constrained low-carbon product swaps, and AI-generated explanations. CAC differs from earlier work by introducing six novel metrics designed for retail checkout involving (i) Carbon Opportunity Gap (COG); (ii) Behavior-Adjusted Emissions (BAE), which accounts for the likelihood that shoppers will accept suggested swaps; (iii) Risk-Adjusted Carbon Score (RACS), which includes uncertainty from life-cycle data; (iv) Basket Marginal Abatement Cost (MAC basket), which shows the cost per unit of emissions avoided; (v) Recurring Purchase Emissions (RPE), which estimates long-term impact; and (vi) Composite Carbon-Health Score (CHCS), which balances carbon reduction with nutritional quality. We built a dataset by merging the Instacart Online Grocery Shopping dataset (3.1 million orders, 50k produces) with product footprint data from Poore and Nemecek, SU-EATABLE LIFE, Open Food Facts, and Eco-Score across 43 food categories. Testing shows a 30.5% emissions reduction with an average price change of ± 1.9%. The system implements all 21 mathematical formulas from our framework and all six metrics. Simulated studies indicate that AI-generated explanations increase swap acceptance by 19 percentage points (36% vs. 17%, p < 0.01) compared to numerical labels alone. The MCP-based design creates auditable records that comply with U.S. Federal Trade Commission Green Guides and proposed EU Green Claims Directive requirements. CAC offers a practical approach to sustainable e-commerce that could enable meaningful decarbonization when deployed at large retailers. CCS CONCEPTS • Information systems → Recommender systems • Applied computing → Environmental sciences • Computing methodologies → Machine learning • Human-centered computing → Ubiquitous and mobile computing