Augmented reality based navigation in liver surgery requires real-time, non-rigid registration of the virtual organ model to the dynamically deforming intra-operative surface. However, computational efforts of the registration process are substantial, which motivates forecasting future deformations to minimize latency and delayed visualization. We propose a novel method to forecast liver deformations in a per-vertex manner. To meet real-time requirements, we investigate exponential smoothing to forecast anchor vertices of the organ model and propagate results to the remaining vertices. Based on a physics simulation, multiple organ model deformation series are generated to study the impact of force area, force magnitude, force direction, and the number of anchor vertices. We achieve forecasting errors in sub-millimeter precision using a Triple exponential smoothing model with 1% anchor vertices, resulting in a computational cost of few milliseconds on standard desktop hardware. This significantly reduces latency and enables increasing process efficiency.