Subject: Information System Science
Title: Skin deep: generative AI vs. rule-based personalization in beauty E-Commerce
Abstract:
The rapid diffusion of Generative AI across digital commerce has created new imperatives for beauty brands seeking to deliver genuinely individualized consumer experiences. This thesis examines how the adoption of Generative AI-powered personalization tools impacts customer experience, perceived personalization, conversion potential, and brand value in beauty e-commerce, with a particular focus on Erborian, a FrancoKorean prestige cosmetics brand and part of the L’Occitane Group, as an empirical case. The study is motivated by a significant gap in the existing literature: while the theoretical foundations of AI-driven personalization, consumer trust, and digital customer experience are well established, empirical research comparing rule-based and GenAI diagnostic tools in a real-world, consumer-facing beauty e-commerce context remains scarce.
The study adopts a qualitative research design grounded in an interpretivist epistemological stance. Primary data was collected through nine protocol-based semi-structured interviews conducted in May 2026 with female cosmetic consumers aged 18 to 30. Each session combined embedded live user testing of two skin diagnostic tools, Erborian’s rule-based text quiz and Yepoda’s Vision AI-powered Skin Analyzer (operated by Haut.AI, trained on three million facial images to assess 150+ skin biomarkers), with a post-experience semi-structured interview exploring six theoretically-derived themes. Data were analysed using a theory-driven thematic analysis (Braun & Clarke, 2006) anchored in four theoretical pillars: the Technology Acceptance Model (Davis, 1989), consumer trust (McKnight et al., 2002; Komiak & Benbasat, 2006), the Personalization-Privacy Paradox (Awad & Krishnan, 2006), and customer experience quality (Verhoef et al., 2009).
The findings reveal a central paradox: Yepoda’s Vision AI diagnostic is perceived as more objectively precise, yet Erborian’s rule-based quiz produces recommendations experienced as more personally relevant. This precision-relevance gap arises from a fundamental asymmetry in information architecture, the quiz captures the consumer’s subjective goal-state, while the Vision AI captures her objective skin-state, and suggests that genuine hyper-personalization requires the integration of both. All nine participants independently converged on the same consumer-derived ideal: a hybrid model combining optional Vision AI facial analysis with structured contextual questioning. The study further finds that brand trust remains the dominant mediator of purchase intention, overriding diagnostic quality in the majority of cases, and that AI disclosure generates a sharply segmented response, 5/9 participants indifferent or positively impacted, 4/9 cautious or resistant, with important implications for transparency strategy design. The Personalization-Privacy Paradox is empirically confirmed and theoretically refined: participants distinguish between AI technology disclosure, which can reduce adoption, and data governance disclosure, which is universally expected and reassuring.
The study makes four theoretical contributions: it introduces the distinction between objective skin-state and subjective goal-state personalization as a refinement of the perceived personalization construct; it demonstrates that brand equity moderates the conversion impact of AI diagnostic tools; it proposes a calibrated transparency model distinguishing between two functionally distinct types of disclosure; and it provides an empirically grounded benchmark for hyper-personalization architecture in beauty e-commerce. For Erborian, the findings yield three strategic recommendations: develop a hybrid diagnostic architecture with an opt-in Vision AI module; anchor the AI narrative in the brand’s authentic K-Beauty tech heritage; and implement a calibrated transparency strategy that foregrounds data governance while contextualising AI technology. The study’s limitations, including sample homogeneity, constant tool exposure order, and researcher positionality, point toward a rich agenda for future quantitative and longitudinal research.
Keywords: Generative AI, beauty e-commerce, skin diagnostics, personalization, customer experience, brand trust, Vision AI, Personalization-Privacy Paradox, hyper-personalization, Erborian, K-Beauty.