High-throughput experimental screening and machine-learning algorithms are implemented in a synergic workflow to predict the photocurrent phase space of organic photovoltaic blends. We identify accurate models employing only the materials band gaps.