In this talk, I will discuss data-driven approaches to reduce computational costs in Fluid-Structure Interaction (FSI) problems through non-intrusive Model Order Reduction (MOR) techniques. I will present three main methodological contributions. First, we explore Dynamic Mode Decomposition (DMD) approaches for non-intrusive Reduced Order Models (ROMs) in FSI problems, achieving significant computational speedups while maintaining good prediction accuracy near steady-state conditions. Second, we develop a novel coupling strategy between a data-driven ROM for solid mechanics and a high-fidelity Full Order Model (FOM) for fluid dynamics, demonstrating accurate predictions even in extrapolation regimes. Third, we introduce an innovative data-driven predictor designed to accelerate the convergence of unsteady partitioned FSI coupling, incorporating an adaptive strategy for robust extrapolation through efficient model retraining on high-fidelity data generated online.