When developing a new vehicle, car manufacturers need to ensure its compliance with strict safety requirements. Aiming to support the engineers in the early phase of this process, we propose a point-cloud neural network approach to enhance transfer learning (TL) for crashworthiness. This work explores the possibility to use the point-based technique to infer knowledge on future situations by exploiting data from past studies. Early on in the development, assessing the crash safety implies dealing with the challenge of low data availability. The engineers need to draw upon their expertise, as the new product can be viewed as a broad modification of past ones. An attractive task is the development of a machine learning approach that learns from the past products and transfers the acquired knowledge to the new ones. TL can achieve this goal. With TL, one learns the basic knowledge from a domain A, also called source domain, and transfers it to a domain B, called the target one, that is characterized by low data availability. The capability of the TL model to distinguish between the two domains is a determining factor in its success. An interesting approach is to include in the learning process the information about the geometry of the considered designs. We propose how to improve the transfer of knowledge by allowing the model to learn the topological differences between A and B. A point-based neural network is merged with the TL theory, and is applied to an explicatory industrial crash example. The previous component designs constitute the source domain; the new component design is considered the target domain, as only a few FE simulations are accessible. The PointNet architecture is modified to three aims: handle the geometrical information of the two designs, learn from the old one, and give predictions on the new one. The obtained results confirm the potential of geometrically-enhanced TL. The proposed methodology can be seen as an innovative solution to automatize the decision making process and, consequently, to improve performance and productivity in crashworthiness development.