Experimental approaches such as randomised controlled trials have been successful in identifying and testing medical technologies. There is enthusiasm for use of similar approaches for health systems policies and programmes. This enthusiasm is a current manifestation of the attempt to model social science on physical science. It fails to appreciate the social nature of health system intervention where everything depends on how people interpret and implement policy, and users respond to new programmes and services. Health systems research needs to build more effectively on thinking from social science. This paper suggests a model through which to conceptualise the health systems research problem, identifies some methods that are consistent with studying its inherent complexity and shows, using a case study, how this approach can inform policy.