Traditional research in service composition has assumed perfect functional matching of service capabilities against stated requirements. In real life, however, this is a myth, as borne out within several SOA development and deployment organizations such as IBM in customer engagements. In particular, the variations in data, functional and nonfunctional requirements present a serious hurdle in reusing existing available services and creating service compositions at run-time. Current research in semantic Web services [1] seeks to address this problem by creating meta-models that capture the domain and later on grounding the requirements and capabilities to this meta-model. However, this is not enough. For example, consider the problem of selecting a flight booking service for an airline. The data, functional and non-functional requirements of this service requirement would probably match the capabilities as advertised by an existing flight booking service. However, if the price adjustment scheme requirement for this airline is different from that available in the flight booking service, then this service cannot be used. In other words, the preconditions under which the flight booking service works could be quite different from those required by the airline. In actuality, the mismatch could be much worse - the data and functional requirements could also end up not matching. (Indeed, we will be highlighting this later in Section 2, via a running example in the insurance domain.) This situation illustrates the importance of modeling and studying variations in services in order to develop more robust service compositions, and lessen the impact of these variations. Such a model would allow the users to measure the amount of variation and determine if any change is necessary to meet the requirements. For example, a service offering can differ from a new requirement in the following ways: (a) the guarantees of the offering are stronger than what the requirement demands, (b), the guarantees of the offering are weaker than the demands of the requirement, and (c) there is an exact match. These variants can be exhibited at several levels of a service offering - data, functional, non-functional, preconditions or effects.