In order to adjust to changing environments and internal states, self-adaptive systems are enabled to autonomously adjust their behaviour. The motive is to achieve better performance while minimising human effort in setting up and maintaining these systems. Ensuring correct functionality across a system's lifetime has been largely addressed. Optimisation of their performance, however, has received little attention. This paper presents an approach that applies goal modelling and decision making theory to calculate the quality of a system's performance in terms of a given configuration's utility with respect to its current environment. Thereby functionally valid configurations can be evaluated within the self-adaptive loop. The approach increased human players' performances in experiments based on a computer game. These results suggests that utility modelling is a promising approach for optimising the quality of behaviour in self-adaptive system.