The polyurethane industry faces challenges in predicting large-scale fire testing performance, particularly for Spray Polyurethane Foam (SPF) applications. Traditional small-scale fire tests provide valuable data, but there is divergence in expert opinion regarding which parameters best correlate with large-scale outcomes. In this collaborative work, HBS and Citrine Informatics present an integrated methodology combining empirical data with advanced AI-driven modeling to improve fire performance predictions. We have leveraged an AI platform that integrates domain knowledge and active learning to refine formulation strategies while providing key insights on thermal degradation patterns. This approach identified strong correlations between small and large-scale fire testing, providing formulators with a more reliable and efficient framework for optimizing fire performance. This integrated solution has significantly reduced the number of large-scale fire tests, along with the associated time and costs. By improving predictive accuracy and reducing formulation trial-and-error, this methodology accelerates and reshapes future product development, lowers testing costs, and supports regulatory compliance in an evolving market.