Collective numerical predictions often outperform individual judgments, a phenomenon rooted in the statistical cancellation of errors under independence. However, real-world group prediction tasks frequently involve interdependencies, social influences, and feedback loops. This study examines the wisdom of crowds in such non-independent settings, specifically within the Transfermarkt.de online community. Users debate and predict soccer player market values, which are later validated by actual transfer fees. We hypothesize that prediction performance increases with (1) group discussion intensity and (2) the accuracy of the initial prediction in a thread. Furthermore, we propose that (3) the positive effect of high discussion intensity will be stronger for less accurate initial predictions. Our analysis of German Bundesliga and Italian Serie A transfers from summer 2015 supports all three hypotheses. Findings enhance understanding of group interaction in complex judgment tasks, suggesting that intense discussion can benefit crowd predictions, even where individual independence is compromised.