Service delivery optimization is a challenge for most service organizations. Most organizations tend to allocate tasks to service workers using simple (and often ad-hoc) policies without leveraging explicit optimization techniques for resource allocation. This is, in part, due to the difficulty of modeling the resource allocation problem in the context of the complex social setting in which service delivery occurs as an optimization problem. This is also, in part, due to several practical impediments to deploying traditional optimization technology in an effective fashion in service delivery settings. This paper offers a novel solution which combines agent-based modeling (ABM) and distributed constraint optimization (DCOP). ABM is used to model the social context of service delivery, while the use of DCOP techniques enables us to bring the dynamic knowledge (and insights) residing in service workers to bear on the optimal resource allocation problem without imposing the unrealistic requirement that all service workers continually communicate their local knowledge to a traditional (centralized) optimization solver. The combination of ABM and DCOP enables us to analyze and assess the efficiency gains that might be possible by optimizing resource allocation in a particular service delivery setting. Our empirical evaluation, based on data patterned on those in a large service delivery organization, provides encouraging results.