Welfare state reform is understood to be risky, difficult and still ongoing. As such, there is a need for analytic tools that can aid understanding of how governments are able to overcome the various barriers they face in seeking change, and the challenges of managing complex and hybridised welfare arrangements. This article argues that the policy tools literature provides several promising avenues for doing so. Specifically, it suggests that procedural policy tools - increasingly recognised in policy studies - are often critical for mitigating risk of blame, sustaining coalitions and supporting the ongoing adaptation of social policy systems to changing conditions and information. This is especially critical where policy innovations entail new substantive relationships between states, citizens and markets, as initial assumptions about individual and group behaviour are likely to be flawed. Procedural tools can enable learning and recalibration in the pursuit of public legitimacy for changing arrangements. To demonstrate the argument, this article engages closely with a case study of public pension reform in Sweden in the 1990s. The case study finds that, initially, reformers made use of a politically insulated parliamentary working group to design a new system, one which attempted to automate elements of policymaking and transfer risk from the state to the individual. However, this goal has proved more difficult to realise than imagined, and in the longer-term, the government used procedural tools for the purposes of learning and recalibration, and in doing so has gone some way to re-establishing modified forms of state accountability.