Selecting multiple, interdependent software components to meet complex system requirements is difficult. This paper reports the experimental application of the i * agent-based system approach to model a system architecture in terms of dependencies between components to achieve goals, satisfy soft goals, complete tasks and provide and consume resources. It describes two treatments, at the architecture and component levels, that can be applied to these architecture models to determine important system properties that inform multiple component selection. The modelling approach and treatments are demonstrated throughout using the example of a meeting scheduler system.