In this study, we harness data science to design carbamate esters (CEs) as donors in Ziegler–Natta catalysis. Using a small yet insightful data set of 18 patented CEs, we developed a multivariate linear regression (MLR) model incorporating key electronic and steric descriptors to predict polymerization yields. Rigorous validation demonstrated the model’s robustness and predictive power, enabling its application in the discovery of higher performing CEs. In the initial optimization cycle, the model guided the design of 10 CEs, which were synthesized and tested, successfully confirming the predictions. A second optimization cycle fine-tuned the most promising CE from cycle 1, leading to the discovery of a highly efficient CE with a yield of 108 kg polymer/g catalyst, marking a 30% improvement over the best performing CE in our initial data set. This work underscores the transformative role of data science in industrial catalyst design, offering a powerful alternative to traditional trial-and-error and density functional theory approaches while accelerating innovation.