There is growing demand for use of serial robotic manipulators in machining, as they are more flexible and affordable than conventional CNC machines. Maintaining positional accuracy under load is crucial to ensure adequate process performance. Offline, configuration-based stiffness optimisation can be used to achieve improved stiffness characteristics for a given process. A new stiffness metric is proposed, which aims to predict how well a process will adhere to its positional tolerances under load for a given joint-configuration. The proposed metric is validated for through-drilling using a Kuka LBR iiwa 14 R820 cobot. It is shown that the metric correlates with the lateral controller tracking error (Spearman’s rank correlation coeffcient of 0.93) and the hole diameter (0.81). The correlations are better than those of existing metrics. The controller error can be decreased from a worst case of 380 µm down to 68 µm, a reduction of more than a factor of 5. A flexible software framework for performing these optimisations is also made available.