Artificial intelligence is increasingly integrated into human resource management, expanding from routine administrative tasks to high-stakes decisions such as promotion, performance-based dismissal, surplus staff clearance, and organizational layoffs. Despite expanding applications, existing literature remains fragmented across functional areas without systematically analyzing how judgment authority is allocated between human managers and algorithms. This paper presents a systematic literature review protocol and conceptual framework to examine human-AI collaboration in consequential HR decisions. Following the PRISMA 2020 statement across Scopus and Web of Science, this study adapts the Antecedents, Decision Process, and Outcomes (ADO) framework. The framework organizes technical, organizational, and individual antecedents; maps a dual-stage decision process (diagnostic analysis versus evaluative assessment) across three collaboration models (AI-informed, AI-augmented, and AI-delegated); and evaluates multi-level outcomes across organizational compliance, employee procedural justice, and societal accountability. By establishing decision stakes as the primary organizing variable, this study clarifies the division between algorithmic analysis and human managerial discretion in consequential workplace decisions.