Textual data are an important information source for risk management for business organizations. To effectively recognize, extract, and analyze risk-related statements in textual data, these processes need to be automated. We developed a design framework for firm-specific risk statements guided by previous economic, managerial, and natural language processing research. Four information types (risk impact, risk type, future timing, and uncertainty) were identified as the key requirements for risk recognition systems. A prototype system, AZRisk, was constructed to verify the framework. Evaluation using news sentences from the Wall Street Journal confirmed the design framework. The performance of AZRisk showed promising results for automated risk recognition.