This article presents a novel framework for creating sense-and-respond rules, which allow detecting noteworthy event situations from streams of business incidents and responding to them in near real-time. Focusing on expressiveness as well as manageability, the proposed framework uses a model-driven approach for the rule definition, where the different aspects of a rule are specified in clearly separated, comprehensible sub-models. This includes models for event-type and correlation information, virtual business-object representations, event patterns ('sense') and actions ('respond'), as well as event processing networks. Event patterns are modelled in a visual decision graph from easy-to-understand pieces of pattern-detection logic, and/or from sub-level event patterns. The proposed system has been fully implemented with a service-oriented architecture. The rule model is illustrated with a business case from the workload-automation domain.