As advanced analytics penetrate a wide range of business applications, companies face the challenge of managing analytics-based assets, such as predictive models. Tasks ahead include model selection, scoring and deployment planning. One way to optimize model selection is to tap the combined knowledge of company staff through a "prediction market," a virtual market designed to reveal participants' aggregate wisdom by seeing where people "invest" their money. In the context of predictive-model selection, this paper refers to such devices as predictive-model markets. This paper examines design possibilities for building experimental markets that can ultimately be used to test whether predictive-model markets will improve model selection and deployment. The researchers test two types of incentives for participation: economic and social. Study results indicate that such markets can effectively work using either; a surprising finding is that social incentives did not improve effectiveness when added to economic incentives.