This thesis synthesizes empirical evidence on human, AI, and human-AI collaboration in frontline marketing and services. The research compiles 170 studies covering 63,376 participants and structures the analysis around two models: (1) human vs. AI and (2) human-AI collaboration vs. non-collaboration. Using a combination of random effects meta-analysis and MASEM (meta-analytic structural equation modelling), average effects were estimated on both behavioural and attitudinal outcomes, and mechanisms were tested via agent-related, functional, and relational mediators, and a set of boundary conditions.
Findings indicate that AI and collaboration reliably improve behavioural outcomes, while attitudinal outcomes are more fragile and depend on functional transparency and relational cues. Paper 1 demonstrates that AI outperforms humans on objective, utilitarian tasks where competence signals dominate, while humans retain an edge where warmth, empathy, and identity-relevant judgment are salient. Meanwhile, Paper 2 discusses how collaboration yields additional gains when role division, autonomy, and oversight are calibrated, while benefits weaken when coordination costs, accountability diffusion, or disclosure frictions arise. Three types of mediators transmit these effects, with boundary conditions and design factors (task type, evaluation and performance judgement, learning cues, disclosure, context, etc.) strengthening or weakening the effectiveness of AI and collaboration.
The thesis contributes an integrated framework that clarifies substitution (human vs. AI model) and complementarity (collaboration vs. non-collaboration model) logics, a consolidated typology of mediators and moderators, and comparable estimates across studies. Managerially, the results offer guidance on when to use humans, AI, or collaboration and how to configure disclosure, autonomy, and accountability to support adoption and performance. Methodologically, the thesis procedures for search, coding, effect-size computation, dependence handling, publication-bias diagnostics, and MASEM reporting, and outlines a research agenda for future studies on human-AI interaction in the marketing and management field.
