Administrative data are increasingly used in impact evaluations of complex interventions in children’s social care (CSC), offering advantages such as longitudinal, population-level coverage, and reduced burden on families and practitioners. However, administrative data alone are insufficient to support meaningful and responsible evaluation. Drawing on insights from two webinars hosted by the Children’s Social Care Data User Group, which focused on presentations about uses of administrative data for evaluations, we identify three core challenges that administrative data cannot resolve in isolation: understanding the intervention, creating a valid comparison group, and measuring relevant outcomes. We explore how the absence of detailed programme theory, limited information on intervention implementation, and lack of nuanced outcome measures constrain the interpretability of evaluation findings. We highlight the risks of confounding in quasi-experiments when key contextual variables are missing from administrative datasets. While we acknowledge the value of administrative data, we caution against over-reliance on datasets designed for operational rather than evaluative purposes. We offer recommendations for more robust evaluation practice, including commitments to building programme theory, prospective data collection, capturing lived experiences, and transparent reporting of study limitations. These recommendations align with emerging policy frameworks and evaluation strategies but require adequate time, funding, and interdisciplinary collaboration.