Service composition is becoming a prevalent way to building service-oriented business applications (SOBAs). In an open service environment, the profit of composition (PoC) is a primary concern of building such applications. How to improve the PoC is a significant issue in developing SOBAs but was largely overlooked by current research. Particularly, the modeling and prediction of PoC should play a key role to drive the composition process. In this paper, we focus on how to model and predict PoC in SOBAs. We regard the PoC of a composite service as a function of the quality of service (QoS) attributes, defined in the service level agreement (SLA) between the service and its external partners. Based on the PoC prediction approach, we further propose a profit-driven composition methodology to assist enterprises to make more profit in their SOBAs.