In exploring the adoption of new return policies and the role of augmented reality (AR) in preventing return behaviour, theoretical and empirical evidence highlights the influence of policy design and technology on consumer actions. The Theory of Planned Behaviour (TPB), proposed by Ajzen (1991), explains how individual behaviour is shaped by attitudes, subjective norms, and perceived behavioural control (Figure 1). In contrast, the Technology Acceptance Model (TAM), developed by Davis (1989), explains technology adoption through two primary constructs, Perceived Usefulness (PU) and Perceived Ease of Use (PEOU). A recent study using TPB found that environmental awareness and perceived behavioural control strongly influence sustainable fast-fashion consumption (Magwegwe and Shaik, 2024). Hagger and Hamilton (2024) provided updated longitudinal and meta-analytic evidence confirming the robustness of TPB in predicting intention–behaviour pathways across consumer contexts. Additionally, TPB combined with Expectation-Confirmation and Cognitive Dissonance theories explains young consumers’ online return intentions through attitudes and perceived ease of returning (Das and Kunja, 2024).

In terms of technology adoption, a recent study reported that PU and PEOU significantly influence mobile application adoption, with higher perceived usefulness leading to stronger behavioural intentions (Mehra et al., 2020). Studies on augmented reality (AR) applications further support the TAM framework. AR try-on systems’ PU and PEOU positively affect adoption intention, which in turn reduces mismatch-related returns (Costa et al., 2025). Similarly, AR usage enhances product knowledge and purchase accuracy, with PU and PEOU significantly predicting behavioural intention (Al Hilal, 2023). Extending this evidence, AR-based product visualisation increases perceived clarity and reduces perceived risk, leading to a lower intention to return fashion items (Sarkis et al., 2025). Virtual product interaction reported that AR improves consumers’ confidence in product fit, reducing cognitive uncertainty and return likelihood (Sun et al., 2022). Similarly, recent findings show that AR-enabled shopping environments enhance decision accuracy and reduce post-purchase dissonance, thereby lowering return intentions (Yoo, 2023).

This study adopts a quantitative, explanatory research design to investigate the adoption of new return policies and the mediating role of Augmented Reality (AR) in preventing return behaviour among online shoppers. The approach is primarily deductive, grounded in established behavioural and technology acceptance theories, while also incorporating an inductive element to refine constructs based on preliminary respondent feedback. Data were collected from online consumers with prior experience in product returns and AR-enabled shopping features, as they represent the most relevant segment for analysing return-related decisions. Initially, a pilot study involving 30 participants was executed to assess the clarity and reliability of the questionnaire. Feedback led to the refinement of AR-related items to better capture perceived usefulness and visualisation effectiveness. The pilot analysis confirmed acceptable validity and internal consistency.

For the main study, a structured questionnaire consisting of Likert-scale items was administered using purposive sampling to specifically target consumers familiar with online shopping and return procedures. The final sample comprised 209 respondents, which is adequate for Structural Equation Modelling using Partial Least Squares (SEM-PLS). The final sample size of 209 respondents meets the recommended minimum requirements for PLS-SEM analysis, particularly for theory-driven models with multiple constructs, as PLS-SEM is suitable for complex models with relatively smaller samples. Data analysis includes Confirmatory Factor Analysis (CFA) to validate construct loadings, reliability, and convergent and discriminant validity. Structural model evaluation will be performed using SEM-PLS to test the hypothesised direct and mediating relationships, supported by model fit criteria such as the Standardised Root Mean Square Residual (SRMR) for goodness-of-fit assessment. Additionally, predictive relevance will be assessed through the coefficient of determination (R²) for key endogenous variables, allowing evaluation of model explanatory power. This combined analytical approach ensures rigorous measurement validation and robust hypothesis testing within the proposed conceptual framework.

The items included to assess the latent variable were derived from validated scales and subsequently modified to align with the research context, as illustrated in Table 1. To improve face validity and comprehension, some changes were done. We assembled a varied panel of prominent academics to evaluate the scale’s clarity and face validity. Their suggestions helped us decide on changes.

The final form of the survey instrument was split into two sections, the first of which contained demographic questions, such as ones on gender, age, level of education, monthly income, and frequency of online purchase return. The other section contained the items for measuring return eligibility policy, non-refundable policy, return shipping fee policy, return abuse prevention policy, brand service centre policy, augmented reality and preventing return behaviour (Table 1). All the items were scored using a 5-point Likert scale, with 5 indicating ‘strongly agree’ and 1 ‘strongly disagree’.

A total of 228 self-administered questionnaires were distributed to regular online shoppers with at least one prior return experience, using Qualtrics. Each questionnaire included a cover letter explaining the study’s objectives, providing instructions, and inviting participants to take part. However, 19 questionnaires were discarded because the respondents did not have at least one prior return experience.

Finally, 209 responses were considered for further analysis. The demographical respondents are presented in Table 2. According to the respondents’ demographic profile (Table 2), most online shoppers are 31–40 years old, followed by 18–30, while very few are above 40, showing low online activity among older consumers. Women make up 60.93% of shoppers, while men make up 39.07%. Online shopping is common across all income groups, with shoppers ranging from those earning below ₹25,000 to those earning above ₹100,000, showing a broad and diverse economic mix. Most respondents shop online monthly (35.2%), 30.6% infrequently, 20.4% rarely, and 13.9% regularly, suggesting scheduled and need-based purchases dominate. Nykaa (27.47%), Purplle (25.34%), Amazon (11.05%), Ajio (10.80%), and Flipkart (10.09%) are the most trusted shopping platforms, followed by Myntra (15.26%), indicating great commitment to major e-commerce firms. Fashion and Apparel (31.7%), Beauty & Cosmetics (26.3%), Home and Lifestyle (20.0%), Eyewear and Sunglasses (12.7%), and Jewelry (9.4%) are the most purchased categories, indicating that consumers rely heavily on e-commerce for visually driven and lifestyle-related products.

Return behaviour analysis shows that 83.3% of respondents had returned a purchase, indicating that fit, mismatch, and dissatisfaction are widespread in online buying. Due to severe return policies, 75.9% have avoided buying a product, demonstrating that they discourage purchases and impact platform choice. Finally, 65.7% of respondents use AR (virtual try-on) before buying, demonstrating a considerable reliance on digital tools to visualise things and eliminate ambiguity. 34.3% of respondents do not use AR, potentially due to lack of understanding. The demographic and behavioural responses show a digitally sophisticated, mostly mid-aged consumer base that bases online shopping decisions on platform trust, product category confidence, return policy flexibility and AR technology.

The measurement model assesses the reliability and validity of the constructs with their corresponding items. Acceptance of the model is dependent on three criteria: its internal consistency reliability, its convergent validity, and its discriminant validity (Ferine et al., 2021).

Internal consistency reliability, a form of reliability, examines the consistency between the observed variables in a test. Cronbach’s alpha and composite reliability are the traditional criteria for determining this; nevertheless, composite reliability is preferable because it prioritizes the indicators based on their individual reliability (Hair et al., 2019). Composite reliability and Cronbach’s alpha values range from 0 to 1, and both must exceed the 0.6 thresholds to be considered acceptable (Hair et al., 2019; Ferine et al., 2021). As shown in Table 3, the value of both composite reliability and Cronbach’s alpha surpasses 0.6, thus demonstrating internal consistency reliability.

SPSS software was used to look at the respondents’ profiles, convergent Validity. We used Smart PLS 4 to test (Figure 2) and confirm the hypotheses of the study model. In the past, covariance-based structural equation modelling was the best way to examine at complicated interactions between observable and latent variables. But in the last few years, PLS-SEM has become much more popular (Hair et al., 2021).

Convergent validity appears when different measures of the same construct exhibit a positive correlation with one another. Measures of convergent validity include factor loadings and average variance extracted (AVE), which must be greater than 0.5 (Hair et al., 2021). As shown in Table 3, all the constructs displayed satisfactory convergent validity, with loadings and AVE values exceeding 0.5. During the initial assessment, the Augmented Reality construct did not meet the convergent validity criterion due to an AVE value below 0.50; therefore, item AR5 was removed to improve convergent validity, after which all constructs satisfied the recommended thresholds.

The final aspect is discriminant validity (Table 4), the extent to which a construct is empirically different from other constructs. Following the guidelines of Hair et al. (2021), the bootstrapping of 209 samples was applied. The measurement model failed to meet the requirements in the initial analysis due to its low AVE value. One indicator was deleted before running the measurement model once.

Examining the structural model is the second step in evaluating PLS-SEM results (Fornell and Larcker, 1981). The direct paths from the independent variables to Preventing Return Behaviour were retained to examine partial mediation effects, in line with standard PLS-SEM procedures. When examining a structural model, it is essential to consider its path coefficients and coefficient of determination (R²) for goodness of fit. The researchers assessed the path coefficients of the structured model in line with Hair et al. (2021) using bootstrapping with 209 samples, a two-tailed test, and a significance level of 0.05, as shown in Table 5. The results demonstrate that apart from AR5, all the hypothesized relationships are observed to be statistically significant.

In particular (Table 5), non-refundable policy shows a significant positive association with preventing return behaviour (PRB) (β = 0.138, t = 2.214, p < 0.05). Similarly, Return Eligibility Policy (REP) (β = 0.27, t = 2.198, p < 0.05) and Service Centre Policy (SCP) (β = 0.235, t = 2.125, p < 0.05) are also positively and significantly related to PRB. However, Return Abuse Prevention Policy (RAPP) (β = 0.19, t = 1.061, p > 0.05) and Return Shipping Fee Policy (RSP) (β = 0.14, t = 1.393, p > 0.05) do not show significant effects on PRB. These results indicate that stricter return eligibility conditions, non-refundable rules, and access to service centre support play an important role in reducing return behaviour, while shipping fees and abuse-prevention measures alone do not significantly influence consumers’ return decisions.

Table 6 presents the mediation analysis, indicating that Augmented Reality serves as a significant mediator between all return-policy constructs and Preventing Return Behaviour (PRB). The indirect effects are statistically significant for the Return Abuse Prevention Policy (β = 0.09, t = 3.248, p = 0.001), Service Centre Policy (β = 0.083, t = 3.192, p = 0.002), Return Shipping Fee Policy (β = 0.073, t = 3.041, p = 0.002), Return Eligibility Policy (β = 0.078, t = 3.137, p = 0.002), and the Non-Refundable Policy (β = 0.086, t = 3.185, p = 0.002). Collectively, these results demonstrate that AR exerts a robust and consistent mediating influence, enhancing the effectiveness of return policies in curbing return behaviour.

The mediator AR had a medium GoF of 0.1142, whereas the GoF result for preventing return behaviour (0.101) suggests a small to moderate model fit (Table 7). According to the benchmarks proposed by Wetzels et al. (2009), GoF values of 0.10, 0.25, and 0.36 can be considered small, medium, and large, respectively. These findings imply that the mediator contributes meaningfully to improving the overall fit and that the model demonstrates sufficient explanatory power.

This study offers several theoretical contributions by integrating the Theory of Planned Behaviour (TPB) and the Technology Acceptance Model (TAM) to examine how new return policies and Augmented Reality (AR) influence preventing return behaviour. First, by applying TPB, the study reinforces the idea that consumer attitudes, subjective norms, and perceived behavioural control are critical in shaping return behaviour. The findings demonstrate that clear and structured return policies, such as the Return Eligibility Policy and Non-Refundable Policy, influence consumer perceptions and decision-making, supporting prior evidence that TPB effectively predicts intention behaviour pathways in diverse consumer contexts. This extends TPB literature by showing that policy design can act as a contextual factor influencing behavioural control and reducing unintended or fraudulent returns.

Second, the study advances the application of TAM in the domain of augmented reality by demonstrating that perceived usefulness (PU) and perceived ease of use (PEOU) of AR systems mediate the relationship between return policies and return behaviour. The significant indirect effects observed for all five policies highlight that AR enhances product evaluation, reduces cognitive uncertainty, and improves purchase confidence, which aligns with prior TAM-based research on AR adoption. This dissertation provides empirical support for the argument that technology adoption frameworks like TAM can be extended to understand consumer responses to policy-mediated interventions, bridging policy and technology adoption literature.

Third, this study integrates insights from both TPB and TAM, demonstrating that policy-induced behavioural control and technology-enabled product visualisation jointly influence return intentions. The findings underscore the complementary role of human behavioural factors and technological facilitation in shaping consumer outcomes. Specifically, AR-based visualisation not only enhances perceived control and reduces perceived risk. Virtual product interaction reported that AR improves consumers’ confidence in product fit, reducing cognitive uncertainty and return likelihood (Sun et al., 2022) but it also strengthens the effectiveness of well-structured return policies, contributing to lower post-purchase dissonance and reduced return intentions (Yoo, 2023).

Overall, this research contributes to theory by extending TPB and TAM beyond traditional settings, showing that the combination of policy design and AR adoption can explain variance in consumer return behaviour. It highlights the potential for hybrid theoretical frameworks that integrate behavioural and technological perspectives to better understand and predict consumer decision-making in e-commerce contexts.

This study can help e-commerce platforms, merchants, and policymakers prevent wasteful or fraudulent returns. First, the study emphasizes organized return policies. The Return Eligibility Policy and Non-Refundable Policy directly and indirectly affect consumer behaviour through AR. By clearly explaining product quality standards, time constraints, and documentation requirements, retailers can encourage more thoughtful purchasing decisions and lower return rates. Second, the study underlines how AR improves customer decision-making. Customers gain confidence and eliminate cognitive ambiguity by using AR technologies to envision products, fit, and comprehend features before buying. App developers and e-commerce platforms can include AR-based product visualization and virtual try-on systems in their return policies to reduce mismatch-related refunds and improve consumer happiness.

Third, AR optimizes return shipping and misuse avoidance. Customer evaluation through AR helps transparent return shipping fees and anti-abuse procedures affect consumer behaviour. To avoid impulsive or fraudulent returns, merchants can use policy reminders, AR product previews, or interactive product recommendations. The study concludes with strategic and digital investment advice. E-commerce companies can improve operational efficiency, return costs, and consumer loyalty by combining policy design and digital technology adoption. Augmented reality can help apparel, home appliance, and electronics retailers improve their return policies and create a more sustainable and successful online retail ecosystem.