Validating segmented algorithmic pathways for ethical online retail personalisation

Purpose Algorithmic personalisation in online retailing advances rapidly, yet single-funnel models fail to capture the multiple routes consumers use to form purchase intentions. Integrating the throughput model (TPM) with self-determination theory and online presence, this study validates TPM's six algorithmic decision pathways and examines how shopping motivation (intrinsic/extrinsic) and online presence (social presence/telepresence) jointly shape judgement and purchase intention, and whether latent segments activate distinct route combinations. Design/methodology/approach Survey data from 677 young consumers in Vietnam's beauty and personal care retail sector were analysed in two stages. Partial least squares structural equation modelling (SEM) tested aggregate structural relationships, after which a hybrid latent class and causal mediation analyses (LCA-CMA) approach combined LCA with causal mediation analysis to uncover unobserved consumer segments and evaluate TPM pathway activation within each segment. Findings Motivation and presence reinforce one another and jointly elevate judgement, the strongest proximal driver of purchase intention. LCA-CMA identifies four latent segments: three activate all six TPM pathways (direct, mediated and serial), while one affect-driven segment relies exclusively on the presence-to-motivation route, bypassing deliberative judgement. Judgement operates as a segment-varying gateway where consumer autonomy is supported or eroded, explaining why uniform funnels misrepresent this heterogeneity and where personalisation risks exploiting judgement-bypassing consumers. Originality/value This is the first study to integrate TPM with self-determination theory and online presence to map segmented online purchase decision algorithms. It introduces a replicable hybrid analytics template that exposes heterogeneity hidden by aggregate SEM, reframing the pathway map as a normative design principle for transparent, ethically grounded algorithmic personalisation.