Knowledge spillovers among firms are widely viewed as a key driver of agglomeration and growth, but are difficult to estimate cleanly. We randomly allocated an energy-efficient motor – a “servo'” motor – among leather-goods firms in Dhaka, Bangladesh, and tracked adoption, information flows, beliefs about energy savings, and other variables. We use the difference between actual exposure and expected exposure (from simulated randomization draws) to identify the effect of exposure. We find a robust positive effect of exposure to treated neighbors within a small geographic area (500 meters in our baseline specification) on information flows and adoption. A marginal value of public funds (MVPF) calculation taking learning spillovers into account yields a significantly larger value than one considering only treated firms and suggests that adoption subsidies would be a cost-effective policy intervention.