Each year billions of US-dollars of humanitarian assistance are mobilised in response to man-made emergencies and natural disasters. Yet, rigorous evidence for how best to intervene remains scant. This dearth reflects that rigorous impact evaluations of humanitarian assistance pose major methodological, practical and ethical challenges. While theory-based impact evaluations can crucially inform humanitarian programming, popular methods, such as orthodox RCTs, are less suitable. Instead, factorial designs and quasi-experimental designs can be ethical and robust, answering questions about how to improve the delivery of assistance. We argue that it helps to be prepared, planning impact evaluations before the onset of emergencies.