Many ideas succeed in small trials but weaken considerably at scale. Using early childhood investment as a case study, this paper develops a dynamic microfounded human capital model stylized in the Chicago tradition. The framework features optimizing agents, complementary skill formation, and a policymaker choosing scaling strategies. The model shows that naive extrapolation from pilots systematically overestimates societal impact by overlooking voltage drops: declining benefit-cost profiles due to unrepresentative samples and contexts. Optimal scaling requires option C thinking, a mechanism-based design approach that anticipates these failures through backward induction from real-world implementation constraints. Studies in this special issue enrich the model’s insights.

We formalize a research design to uncover the mechanisms underlying long-term reductions in energy consumption caused by a widely implemented nudge. We consider two channels: technology adoption and habit formation. Using data from 38 natural field experiments, we isolate the role of technology adoption by comparing treatment and control homes after the initial resident moves, which discontinues the treatment for a home. We find that fully half of energy reductions persist in the home after treatment ends and show this persistence is consonant with a technology adoption channel. The role of technology in creating persistent behaviour change has important implications for designing behavioural interventions and evaluating their long-term social impacts.