Low latency interactive data visualizations of large volumes of data are still rare on the web today. Their development requires expertise in server development, API design, dataflow optimization, and browser technologies; developers rarely have the time to optimize the whole stack from server to browser. To lift the burden of client-server co-development, we propose a system that generates visualization applications from declarative visualization and interaction specifications. We envision a system that automatically optimizes a visualization plan to reduce latencies, especially in low-connectivity or mobile networks. In this paper, we investigate the design of automated techniques to determine a partition of work across server and client that minimizes latency. Our cost model is based on a combination of data statistics, network performance, available computing resources, and predicted interactions. As a preliminary evaluation of our approach, we describe how our system could work in the context of two analysis scenarios.