In this project, I integrate machine learning (ML) technology and traditional economic frameworks. The explosion of data and computational power are enabling empirical researchers to uncover more detailed insights. While adoption of ML has been widespread, the fields of economics and policymaking have been hesitant because of their focus on causality rather than predictions. Combining prediction power with causal inference, I aim to offer both private and public institutions with new methods and insights in order to design effective policies using large micro-level data. In particular, I center my research around bridging the gap between out-of-sample prediction (ML) and causal inference (economics). By embedding such methods within specific policy problems, I increase the scope and reach of policy decisions in the context of insurance and fiscal policy. I break down policy analyses into three parts. First is treatment assignment, to which I contribute by investigating the efficiency and fairness of insurance pricing, where assignment, here pricing, is based on a risk prediction. The second contribution adds to (heterogeneous) treatment effect estimation by studying saving behaviour changes induced by the 2020 elimination of tax deductions on mortgages throughout the population in Flanders, Belgium. Lastly, I construct a generative modelling method that preserves causal structure of data which allows to evaluate the generalizability and transportability of treatment effects.