Plastic anisotropy may strongly affect the stress and strain response in metals subjected to multiaxial cyclic loading. This anisotropy evolves due to various microstructural features. We first use simple models to study how such features result in evolving plastic anisotropy. A subsequent analysis of existing distortional hardening models highlights the difference between stress- and strain-driven models. Following this analysis, we conclude that the stress-driven approach is most suitable and propose an improved stress-driven model. It is thermodynamically consistent and guarantees yield surface convexity. Many distortional hardening models in the literature do not fulfill the latter. In contrast, the model proposed in this work has a convex yield surface independent of its parameter values. Experimental results, considering yield surface evolution after large shear strains, are used to assess the model's performance. We carefully analyze the experiments in the finite strain setting, showing how the numerical results can be compared with the experimental results. The new model fits the experimental results significantly better than its predecessor without introducing additional material parameters.