<p>How is consensus reached in groups with limited access to information? Here we run a controlled laboratory experiment with very young children (ages 5–8 years, <i>N</i> = 150) in the USA to identify the attributes that support consensus building. Success is facilitated by two factors. The first is the endogenous adoption of heterogeneous roles: a leader who proposes a solution, a group of debaters who consider the alternatives, and a closer who locks the decision. The second is flexibility in the decision rule, with participants following the wisdom of the crowd with high probability, but not with certainty. These two characteristics allow young children to outperform simple computational algorithms, especially in the more complex conditions with limited observability of the network: children converged 74% of the time, compared with 35% for the algorithm (<i>χ</i>²(1) = 59.83, <i>P</i> &lt; 0.001, 95% confidence interval for the difference in proportions 0.30–0.47). The study also reveals a sharp progression with age (<i>β</i> = 1.49, <i>P</i> &lt; 0.001, 95% confidence interval 0.62–2.36) corresponding to a 4.45-fold increase in the odds of convergence for each one-unit increase in grade. This work contributes to our understanding of how children navigate collective problems in complex social environments.</p>