Forging is a widely used manufacturing process for producing high performance components. Due to its nonlinear behavior, accurately predicting the final geometry of forged parts is challenging. As a result, conservative tolerances and additional operations are commonly employed, leading to increased material losses, additional processing steps, and reduced production sustainability. Material losses are generated from conservative tolerances applied to forged components; improving dimensional precision allows tolerance reduction and consequently decreases material waste. Therefore, minimizing dimensional errors, by achieving closer similarity between the forged part and the target geometry, is a key objective in forging optimization. Traditionally, these issues are addressed through experience based methods. In contrast, structural optimization provides a systematic alternative by formulating the design task as a mathematical problem. Previous studies have addressed this problem by using surface nodes or control points as design variables. An alternative approach is to use geometric parameters as design variables. In industry, feature based parametric CAD is the standard technology for modeling processes, enabling efficient geometry modification through a limited set of parameters. However, this strategy has been rarely explored in this context. This study presents an iterative surrogate based optimization methodology to minimize dimensional errors in the forging process by adjusting die geometric parameters. The proposed computational framework combines nonlinear Finite Element Method simulations with orthogonal experimental design, to reduce the number of simulations required, and iterative surrogate modeling. At each iteration, the parameters that minimize the objective function are calculated, and the resulting optimum guides the selection of new simulation points, expanding the dataset with additional samples around the optimum. Preliminary results indicate that the approach can effectively reduce dimensional errors in the forging process. The proposed approach can be generalized to other optimization problems involving expensive objective function evaluations.