The global environment is changing and policy scholars point to the need for more innovative and information-driven solutions. Sophistications in environmental measurement and modeling may offer opportunities for making environmental policies more transparent and cost-effective, but institutions that oversee policy processes may be slow to change and sometimes resist change. In this dissertation I look at the role of policy networks in conceptualizing the changing nature of environmental governance in the digital age. I study how policy actors develop arguments, mobilize resources, and work around new policy models to resist or control change dynamics. Seen through the perspective of policy networks, the success of new policy ideas depends on how and to what extent incumbent actors are able to interpret, adapt, and absorb changes in their own terms. I focus on innovative policy models such as outcome-based approaches to reform agri-environmental policies in the US. My dissertation centers on an environmental quantification algorithm mobilized to rationalize conservation subsidies in a $10 billion federal US agricultural program. By foregrounding performances of bureaucrats building and using the information infrastructures, I contrast the dynamic potential of data-driven technologies with the rigidity of bureaucracies. I conduct a mixed-methods analysis: historical study of how the algorithm was designed, an ethnography of how the algorithm is used by street-level bureaucrats, and an econometric analysis of public spending on over 55,000 contracts. Drawing attention to the performances of data-driven conservation at different levels of the government opens a critical and timely debate on how information triggers policy innovations. I show that information both disciplines bureaucratic discretion and yet the very legitimacy of the numbers depends on the trust in the bureaucracy’s established culture and routines. The paradoxical dynamic ensures that without a change in the larger political-administrative structures that shape confidence in the calculations, data-driven policies may perform calculation but produce little policy change.