Systems are becoming exceedingly complex to manage. As such, there is an increasing trend towards developing systems that are self-managing. Policy-based infrastructures have been used to provide a limited degree of automation, by associating actions to system-events. In the context of self-managing systems, the existing policy-specification model fails to capture the following: a) The impact of a rule on system behavior (behavior implication). This is required for automated decision-making. b) Learning mechanisms for refining the invocation heuristics by monitoring the impact of rules. This paper proposes ‘Eos’, an approach to enhance the existing policy-based model with behavior implications. The paper gives details of the following aspects: 1) expressing behavior implications; 2) using behavior implications of a rule for learning and automated decision-making; 3) enhancing existing policy-based infrastructures to support self-management using Eos. The paper also describes an example of using Eos for self-management within a distributed file-system.