In recent years, the integration of machine learning (ML) into catalysis and organic synthesis has attracted interdisciplinary interest. However, because catalytic systems encompass highly diverse reaction types and activation mechanisms, the optimal implementation of ML can vary substantially from one system to another. Consequently, the development of tailored ML strategies suited to specific catalytic contexts is essential. In this article, we focus on organic photosensitizers—molecular catalysts that enable a variety of synthetically valuable transformations through photoredox and energy-transfer processes. From a data-science perspective, the structural diversity of these catalysts, the broad range of reactions they mediate, and the inherent complexity of photoactivation mechanisms provide a compelling platform for the development and evaluation of ML strategies. We primarily introduce transfer-learning (TL) strategies as a promising approach to address the data scarcity in organic synthesis. Specifically, we describe TL methodologies that leverage data from distinct but related reactions, as well as information transfer from virtual molecular databases to real-world organic photosensitizers, and their application to efficient catalyst screening. In addition, we discuss an ML-based strategy to elucidate correlations between elusive excited-state properties and photocatalytic activity through the integration of quantum chemical calculations and data-driven modeling.