One of the most challenging aspects in applying decentralized detection in sensor networks is the efficient exchange of small messages required for data fusion. In this work, we propose a novel communication architecture for a canonical decentralized detection problem where the sensor nodes exchange continuously their local decisions until consensus is reached among all nodes. Our methodology capitalizes on the observation that the information embedded in the exchanged messages decreases to zero as the decisions gradually converge. By using a data-driven multiple access scheme, we show that the number of channel accesses required for each round of message exchange decreases, following the same trend as the aggregate entropy of the sensor decisions. The main contribution is to show that data-driven multiple access strategies can overcome the backlog of communications that many distributed computing algorithms generate in a wireless network setting.