Purpose:　This　study　aims　to　investigate　how　service　agent　type　(human　vs　robot)　influences　consumers’　punishment　intentions　following　service　failure.　Moreover,　it　examines　the　mediating　mechanism　of　anticipatory　guilt　and　the　boundary　condition　imposed　by　consumer　relationship　orientation　(exchange　vs　communal).　Design/methodology/approach:　Four　online　experiments　(N　=　978)　systematically　test　the　hypotheses:　Study　1　uses　an　experience　recall　task　to　establish　baseline　effects.　Study　2　uses　a　controlled　experiment　comparing　human　and　robot　service　failure　scenarios.　Study　3　probes　the　mediating　role　of　anticipatory　guilt　through　a　moderated　causal　chain　analysis.　Study　4　assesses　the　moderating　effect　of　relationship　orientation　using　a　2　(agent　type)　×　2　(orientation)　factorial　design.　Findings:　The　results　reveal　that　consumers　exhibit　significantly　lower　punishment　intentions　toward　human　service　agents　than　toward　their　robotic　counterparts,　primarily　because　of　greater　anticipatory　guilt.　This　effect　is　particularly　pronounced　among　consumers　with　communal　(vs　exchange)　relationship　orientations.　Alternative　explanations　based　on　perceived　warmth,　competence　and　retaliation　risk　are　empirically　ruled　out,　demonstrating　the　robustness　of　the　findings.　Originality/value:　This　research　extends　service　failure　scholarship　by　investigating　how　consumers’　advanced　awareness　of　punitive　outcomes　influences　their　punishment　intentions　toward　service　agents　(robots　and　human　employees).　While　the　literature　predominantly　focuses　on　dimensions　of　robot　agency　as　conceptual　mechanisms,　this　research　pioneers　an　affective　mediation　framework,　using　empirical　validation　to　establish　anticipatory　guilt　as　the　critical　psychological　mediator.　This　research　also　demonstrates　how　relationship　orientations　(exchange　and　communal)　modulate　such　effects,　thereby　advancing　human–computer　interaction　studies　through　an　emotional　lens.　©　2025,　Emerald　Publishing　Limited.