Transition metals are ever-present as reactive centers in biological and inorganic catalytic cycles. However, the open-shell character that gives 3d transition metals unique reactive properties also makes transition metal complexes challenging to study using first-principles approaches, especially when using local or semilocal approximations to density functional theory (DFT). We describe here an approach based on Hubbard U corrections—widely used in the solid-state community to describe strongly correlated systems—and show how it helps achieve predictive accuracy in DFT calculations of transition metal complexes. The success of this approach comes from counteracting the tendency of 3d electrons to delocalize, driven by the imperfect cancellation of electrostatic self-interactions in common exchange correlation approximations. Since the Hubbard term U is calculated through a linear response formulation—recently extended by us to allow for self-consistency—it represents a fully ab initio, nonempirical approach. We analyze the performance of the DFT+U formulations on a few paradigmatic test cases, with special attention to the structure, electronic structure, and potential energy surface of Fe2 dimers and the addition–elimination reaction of hydrogen or methane on FeO+. Thanks to negligible computational overheads, we also show how the approach can be effortlessly applied to large-scale simulations, such as the case presented here of functionalized cobalt porphyrins on a metal support.