Generative AI-based robo-advisors (GARA) offer transformative phygital financial experiences but face trust issues due to algorithmic opacity. Drawing upon cognitive load theory (CLT), this study examines the cognitive and affective mechanisms through which algorithmic transparency (AT) shapes users' financial well-being (FWB). Reconceptualizing AT as a cognitive scaffold that reduces extraneous cognitive load (CL) rather than as mere information disclosure, the study tests a dual-path model linking AT to FWB via perceived utility (PU) and psychological comfort (PC), and examines the moderating roles of financial risk tolerance (FRT) and suspicion toward human advisors (STHA).
Data were collected from 501 experienced robo-advisor users in the United States (n = 251) and South Korea (n = 250) through an online survey. Hypotheses were tested using partial least squares structural equation modeling (PLS-SEM), complemented by bootstrapped mediation analyses, continuous moderation analyses for FRT and STHA, and a multi-group analysis (MGA) to examine cross-national differences.
AT enhances FWB both directly and indirectly by reducing CL, which in turn strengthens both PU and PC. However, only PC significantly predicts FWB, whereas PU does not. Consistent with this asymmetry, the sequential mediation path AT→CL→PC→FWB is supported, while the cognitive counterpart via PU is not. Moderation analyses indicate that FRT conditions the CL→PU and CL→PC links, whereas STHA conditions the AT→FWB, CL→PU, and CL→PC links. The MGA further reveals that the affective pathway (PC→FWB) is significantly stronger in South Korea than in the United States.
This study extends CLT to AI-mediated financial services and establishes that AT operates on FWB primarily through an affective rather than cognitive route, repositioning PC – not PU – as the focal mediator linking transparency to well-being. In doing so, it shifts the evaluative lens of Explainable AI (XAI) research from comprehension accuracy toward emotional regulation, and identifies boundary conditions shaped by individual differences (FRT, STHA) and cultural context (USA vs. South Korea). The findings offer strategic guidance for designing responsible, user-centric GARA systems that integrate efficiency, transparency, and affective reassurance.
