In this paper, we focus on a policy trial that was undertaken in a complex and controversial policy field (Indigenous community development), with the avowed objective of learning from experience in the field . We observe that despite significant implementation effort and quality evaluation, little systematic policy learning occurred. Four impediments are identified: (i) the policy environment was highly politicized and contested, (ii) the results challenged accepted ways of working, (iii) there were no systematic processes for policy learning to occur, and (iv) the learnings were distorted by inter‐agency competition and associated ‘labelling’. We suggest that, in controversial policy fields, the most useful first step in enhancing evaluation‐use is for bureaucracies to institute systematic processes for evaluation findings to be considered, with the aim of elaborating ‘what works’ into a consensual approach to change. Without change of this kind, evaluation‐use will remain a failed area of policy.