Why Older Consumers Resist Voice Commerce: An Extended UTAUT2 Model of Techno-Anxiety, Privacy Calculus, and Trust - SHAP Evidence from Urban India

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Purpose: Voice commerce is proliferating at a higher rate than most other retail product categories, although it remains significantly under-adhered among older consumers in India. The current research explores adoption barriers to the use of voice commerce among Indian adult citizens aged 55 and above, who reside in six metro cities. An extended UTAUT2 model is applied to the data collected, so that techno-anxiety, voice accuracy perception, privacy concern, and trust in voice AI are included in the framework as additional predictors of adoption intention and usage behavior.Design/Methodology/Approach: The cross-sectional survey of the 412 respondents with a mean age of 61.4 was analysed in two successive stages. The first stage used SmartPLS 4.0 software to perform partial least squares structural equation modelling analysis. The second stage applied Random Forest and XGBoost classification algorithms with the SHAP (SHapley Additive exPlanations) interpretation to identify the SEM variables that were misestimated (non-linear effect) and to cross-validate the results of the SEM analysis.Findings: Techno-anxiety is the dominant adoption barrier (β = −0.298, p < 0.001, f² = 0.148), exceeding privacy concern (β = −0.224) and all four UTAUT2 base construct effects. Voice accuracy perception is the strongest positive predictor beyond the UTAUT2 core (β = 0.241, p < 0.001). XGBoost achieves 86.1% classification accuracy (AUC-ROC = 0.907). SHAP analysis ranks voice accuracy perception second in feature importance - above its PLS-SEM β rank - revealing a threshold effect that the linear model does not capture. Age cohort (55–64 versus 65 and above) significantly moderates the adoption intention to actual use path.Originality/Value: This study is the first to treat voice-interface techno-anxiety as a distinct construct in an UTAUT2 voice adoption framework and to validate voice accuracy perception as a novel predictor. The PLS-SEM and SHAP dual design identifies where non-linear predictive contributions diverge from linear path coefficients, offering a replicable methodological approach for adoption researchers. The findings shift design priority for voice commerce platforms from feature development toward anxiety reduction and speech recognition accuracy for older populations.