AI Personalization and Third-Party Product Sales in Delhi NCR Banks: A PLS-SEM Analysis

Abstract

AI personalization in Banking, specifically for third party products, has become increasingly prevalent. This research paper studies the research topic as part of an ongoing PRISMA guided systematic literature review. A conceptual model was developed using the Personalization – Privacy Paradox Theory which included five constructs. The five constructs were AI Personalization Effectiveness (APE), Algorithmic Trust Propensity (ATP), Perceived Security & Privacy Assurance (PSPA), Regulatory Awareness & Sensitivity (RAS) and Third-Party Product Sales Outcome (TPSO). Hypotheses testing was conducted using PLS-SEM method and SmartPLS 4.0 software. Primary data was obtained from 350 customers within Delhi NCR who have experienced AI driven banking services. All exogenous constructs were found to have a significant positive relationship with TPSO; however, RAS was identified to be the strongest mediator and therefore had the most direct influence on TPSO (β = 0.498; p<0.001). Further mediation analysis revealed that RAS significantly mediated both ATP & PSPA to TPSO whilst APE's influence on TPSO was found to occur mainly directly. Overall, the proposed model accounted for approximately 61.3% of the variance in TPSO (R² = 0.613). In addition, robustness checks demonstrated that removing PSPA or RAS from the full model resulted in no loss in explanatory power or prediction accuracy demonstrating the incremental explanatory and predictive value of the regulatory mediation path. Finally, the current study contributes to the personalization-privacy paradox by providing empirical evidence of an ethical dimension, which has previously not been operationalized in India’s banking literature for AI compliant regulation. The novelty of this study lies in its focus on third party products sold via AI driven processes and banking services provided by banks operating in Delhi NCR. Therefore, it should be viewed strictly within these specific methodological and contextual boundaries. The limitations of the study include a cross-sectional study design, limited geographic area (only Delhi NCR), purposeful sampling of digitally active customers and aggregating results across three product categories. Future studies could utilize longitudinal designs, increase their geographic coverage, perform multiple group comparisons and investigate moderating variables.