We study a fully connected network (cluster) of interacting two-state units as a model of cooperative decision making. Each unit in isolation generates a Poisson process with rate g . We show that when the number of nodes is finite, the decision-making process becomes intermittent. The decision-time distribution density is characterized by inverse power-law behavior with index mu=1.5 and is exponentially truncated. We find that the condition of perfect consensus is recovered by means of a fat tail that becomes more and more extended with increasing number of nodes N . The intermittent dynamics of the global variable are described by the motion of a particle in a double well potential. The particle spends a portion of the total time tau(S) at the top of the potential barrier. Using theoretical and numerical arguments it is proved that tau(S) is proportional to (1/g)ln(const x N) . The second portion of its time, tau(K), is spent by the particle at the bottom of the potential well and it is given by tau(K)=(1/g)exp(const x N) . We show that the time tau(K) is responsible for the Kramers fat tail. This generates a stronger ergodicity breakdown than that generated by the inverse power law without truncation. We establish that the condition of partial consensus can be transmitted from one cluster to another provided that both networks are in a cooperative condition. No significant information transmission is possible if one of the two networks is not yet self-organized. We find that partitioning a large network into a set of smaller interacting clusters has the effect of converting the fat Kramers tail into an inverse power law with mu=1.5 .