This study examines the impact of interactional justice -informational justice and interpersonal justice-on customer service evaluation in banking. Through five scenario-based experiments with 921 Chinese non-banking adults, we systematically manipulated justice levels and service agent source (human vs. AI). Findings reveal that both dimensions significantly enhance service evaluations (supported by a mini meta-analysis: Hedges' g = 0.62 for informational, g = 1.16 for interpersonal justice). Crucially, we identify a critical theoretical boundary condition for algorithm aversion: under low justice conditions, human agents received significantly lower evaluations than AI agents, yet under high justice, no significant human-AI difference emerged . This asymmetry-driven by higher expectations for humans-challenges the universality of human preference in service recovery and demonstrates that AI can achieve parity when justice is optimized. Our results advance interactional justice theory and provide actionable insights for resource allocation in AI-human hybrid service systems.