Wireless sensor networks are often deployed for monitoring purposes: when nodes detect the occurrence of a significant event, they transmit an information to a control entity in a multi-hop fashion. When data rate increases, congestion becomes a fundamental issue, especially when an emergency situation generates alarm messages originating from a specific area of the network. Indeed, congestion increases delays and packet losses, and yields to an unfair use of the energy of nodes. In this paper, we propose to improve the ZigBee routing protocol, aiming at reducing traffic congestion. The proposed solution uses intermediate nodes, denoted as pivots, which are selected by the data sources in order to reduce congestion on paths. Simulation results highlight the significant improvement achieved in terms of packet losses and average delays, with respect to the ZigBee routing protocol, while the overhead generated in the network is maintained under control. A mathematical model to derive the average path length and the number of pivots is also provided.