Industrial robots have been around since the 1960s and their introduction into the manufacturing industry has helped in automating otherwise repetitive and unsafe tasks, while also increasing the performance and productivity for the companies that adopted the technology. As the majority of industrial robotic arms are deployed in repetitive tasks, the pose accuracy is much less of a key driver for the majority of consumers (e.g. the automotive industry) than speed, payload, energy efficiency and unit cost. Consequently, manufacturers of industrial robots often quote repeatability as an indication of performance whilst the pose accuracy remains comparatively poor. Due to their lack in accuracy, robotic arms have seen slower adoption in the aerospace industry where high accuracy is of utmost importance. However if their accuracy could be improved, robots offer significant advantages, being comparatively inexpensive and more flexible than bespoke automation. Extensive research has been conducted in the area of improving robotic accuracy through re-calibration of the kinematic model. This approach is often highly complex, and seeks to optimise performance over the whole working volume or a portion thereof, rather than optimising performance of a particular task. In this paper, a method for iteratively teaching poses on a standard industrial robot is presented, and an investigation into the limits on the achievable pose accuracy and the required recalibration period is conducted. Through experimental work on a KUKA KR 240 R2900 ultra robot equipped with a drilling end-effector and measured in 3DoF using a laser tracker, it is demonstrated that the achievable accuracy approaches the stated repeatability of the robot. Finally, investigation results into the accuracy of the robot over short distances to allow small corrections to be applied from these taught poses to compensate for work-piece alignment or thermal effects are presented.