Vehicles are important assets generating significant costs. Relief organizations frequently struggle to define the appropriate number of vehicles to support their operations. The high level of decentralization giving country offices the autonomy to decide on the size of their fleet complicates the issue. This study follows a design science approach in collaboration with the Office of the UN High Commissioner for Refugees (UNHCR). We develop a prediction model to support UNHCR's fleet sizing problem. A stepwise linear regression approach is used to construct a model able to predict the number of vehicles required by each country office based on data from comparable countries. Three variables have the best predictive accuracy: the number of locations, small partners, and large partners working for UNHCR. We validate our findings with different regression methods and by applying our approach to another organization. Our model has provided UNHCR with valuable indications on how to help determine the appropriate number of vehicles in many countries. We develop three design propositions that show how our approach can be generalized to other humanitarian operations. These propositions offer insights on how to implement a fleet sizing decision process in highly decentralized humanitarian operations with limited information on optimal fleet sizes.