Computational simulations have become an important tool for the design of mechanical circulatory support devices as well as surgical planning ahead of implantation. While computational fluid dynamics can accurately predict flow fields, the prediction of hemolysis (red blood cell damage) remains challenging. Absolute predictions of hemolysis indices often deviate by multiple orders of magnitude from experimental measurements. This discrepancy can be attributed to two issues. First, most existing models for hemolysis employ a simple power law relationship between shear stress, exposure time, and hemolysis index, with model parameters fitted to experimental data. Second, experimental data often exhibits large variability between donors and between studies, leading to significant uncertainty in the fitted model parameters. This is due to individual differences in red blood cell properties and high sensitivity to experimental conditions. Consequently, the predictive capabilities of these models are limited, especially when applied to flow conditions that differ from those used in experiments. We propose a two-sided approach to enhance the predictive capabilities of hemolysis models. First, we introduce a more physiological model for hemolysis that incorporates two important effects of the red blood cell membrane: viscoelastic deformation and pore formation. Second, we show how uncertainty quantification techniques can be employed to account for the variability in experimental data when fitting model parameters. We provide confidence intervals for the predicted hemolysis indices. We demonstrate the effectiveness of our approach through numerical simulations of blood flow in FDA benchmark geometries. Overall, this two-sided approach allows us to analyze the impact of modeling assumptions and experimental variability on hemolysis predictions. Further, it facilitates the integration of new experimental data as it becomes available, thereby enabling patient-specific predictions of hemolysis.