The 21st century faces an accelerating climate change challenge, requiring deep, fast and sustainable reductions in greenhouse gas emissions. Among them, Scope 3 emissions covering indirect emissions across a company’s entire value chain are critical but difficult to estimate due to scarce and unreliable data. This paper proposes a new empirical methodology to estimate firm-level Scope 3 emissions by integrating value chain dynamics and company-specific characteristics. Using input–output tables and sectoral emissions data, we reconstruct company value chains to capture upstream and downstream emissions. To address missing data, we apply both parametric models and machine learning techniques to estimate reported and unreported emissions. Using French firm-level data, our results suggest that company characteristics and sectoral emissions throughout the value chain strongly influence Scope 3 emissions. Machine learning models, particularly Random Forests, significantly outperform traditional models. Overall, our findings highlight the importance of improved emissions reporting and comprehensive climate policies to better manage emissions across all sectors.