By explicitly modeling the decision making of heterogeneous individuals, agent-based models can compute the resulting emergent phenomena on the micro-level. This lets planners evaluate new planning approaches for problems that depend on individual decisions. Examples include airline revenue management or traffic control. However, when decision support relies on agent-based modeling, its applicability to real-world problems depends on the model's validity. This paper introduces a novel methodological concept to decompose agent-based models for calibration and validation. This concept enables modelers to isolate agents from the evolution of the model's state variables, allowing greater choice of calibration and validation approaches. The approach first parameterizes and validates individual agents, and subsequently re-calibrates the agent-collective within the entire model.