Companies invest substantial capital in advertising campaigns for products and services. Advertisements that generate undue controversies can completely destroy an advertising campaign. Given the large investments and the explosively viral nature of the spread of controversies, early detection of potential controversies is of vital importance in deciding the future course of these campaigns. However, it is difficult to estimate the potential of controversies through traditional methods such as customer surveys and market research. In this paper, we develop a controversy detection system based on initial comments on online advertisements posted on YouTube. We extract early YouTube comments on a collection of around 45 Superbowl advertisements. We generate a comprehensive set of over 2500 semantic and linguistic features and evaluate their efficacy in automatically detecting controversial comments. Our results show good accuracy in early detection of controversies. The proposed data-driven approach can complement and greatly aid traditional approaches of market research.