Crises are destabilising, and crisis exercises are one of the core ways that organisations and governments prepare. Not all crisis exercises achieve their goals, and determining how to design them well is difficult to test experimentally. This study aims to identify i) the key goals of discussion-based crisis exercises, ii) what features of design and delivery are important for achieving the aims of discussion-based exercises, and iii) common mistakes. We used a three-round Delphi Method to harness the collective expertise of a diverse range of 16 researchers and professionals. In the first round, experts identified 40 different exercise features. After a further two rounds, consensus was achieved for 29 of the 40 exercise features. Overall, experts agreed on the critical importance of clear objectives, good facilitation, and quality debriefing. Experts also agreed that features such as fidelity and the explicit use of decision-making tools were not essential to achieving the aims of discussion-based exercises. There was disagreement amongst experts with regard to how much say the sponsor should have, how actively involved facilitators should be, and how accountable responders should be held. Overall, these findings are beneficial for exercise designers in highlighting what features to prioritise when designing and delivering discussion-based exercises to best promote organisational learning.