Climate services are intended to improve climate-sensitive decisions by making climate information ‘useful, useable and used’. Here, we analyse 27 expert interviews to evaluate whether this user-driven model of climate science has been successfully implemented in the public sector. We show that, although climate services promise better decision-making, they mainly focus on delivering better data. The norms and institutions of climate science produce three key tensions in operationalizing climate services: a focus on products rather than processes, services based on broad assumptions about demand rather than being demand-driven, and the narrow economic valuation of products rather than evaluation of improvements in decision-making. These tensions help explain why climate services often generate nominal changes in climate science where transformations are promised. Transformational change requires that climate services account for diverse social structures, behaviours and contexts. Integrating social science is no panacea for demand-driven climate services, but it is certainly a prerequisite.