Behavioural drivers of internet of things adoption in food consumption

Industrial Revolution 4.0 enables the acceleration of the development and modernization of the food supply chain. Technologies such as artificial intelligence (AI), machine learning (ML), the Internet of Things (IoT), big data analysis, and blockchain are revolutionizing the entire agri-food sector, with positive impacts on environmental sustainability. The analysis of IoT data streams using AI provides timely, evidence-based insights that facilitate real-time monitoring and optimised decision-making (Frau et al., 2022). IoT is the data layer, supplying continuous real-time signals, and AI/ML is the intelligence layer, utilising this data to make predictions, recommendations, and automated actions (Dakhia et al., 2025). These modern technologies are designed to address the challenges posed by the growing global demand for food, while also tackling the issues surrounding unsustainable food production, distribution, and consumption (Raj et al., 2021). Consequently, AI-enabled digitalisation increases efficiency and productivity while improving environmental sustainability for farmers, consumers, and society. Recent research shows that integrating IoT with blockchain solutions can further enhance efficiency and substantially reduce food waste across the supply chain, underscoring the transformative potential of digital technologies in sustainable food management (Zhang et al., 2025). Our study does not empirically examine the mechanisms of interaction between the IoT and AI; instead, we treat AI/ML as a context-aware analytical layer that can leverage IoT-generated data.

The transformation of the food market depends on the rate of IoT technology adoption and the establishment of trust among end users. Therefore, it is reasonable to explore both the technical and psychological factors influencing the acceptance or rejection of IoT in the context of food consumption. IoT is still in its early developmental phase with limited market penetration, facing challenges such as privacy and security concerns regarding processed data (Jerzyk et al., 2024; Kamble et al., 2019), which increases consumer skepticism, in turn, shaping attitude formation and behavioural intention (Rigopoulou et al., 2015).

Despite the growing interest in IoT, research on consumer use of this technology in the food market remains limited. Typically, studies focus on issues such as food deliveries and ordering (Lee et al., 2023), drone deliveries, food application usage, product tracking systems (Bandinelli et al., 2023), and self-service checkouts for restaurant orders (Jeon et al., 2020). A frequently discussed topic in research is the acceptance of modern food production technologies (Giacalone and Jaeger, 2023; Gómez-Llorente et al., 2022). In our research, we focus on identifying and understanding the factors that influence consumers' adoption of innovative technologies in food consumption. This is important insofar as the acceptance of IoT in the area of food consumption provides the opportunity to choose products that generate a lower carbon footprint and come from sustainable practices (Kamble et al., 2019; Wolfert et al., 2017), which, in the long term, puts market pressure on producers to use more sustainable production methods, reducing the consumption of natural resources and negative environmental impacts (Caferra et al., 2023; Yamoah et al., 2022). IoT solutions in the food market have the potential to better meet consumers' food demands while also supporting sustainable economic development. However, the implementation of IoT-based solutions, alongside technological and organizational challenges, poses numerous threats at the consumer level, reducing behavioural intention and delaying the adoption of IoT in the food market (Gbashi and Njobeh, 2024; Jerzyk et al., 2024).

Researchers have long been interested in how innovations are adopted and diffused. Various theoretical models have been proposed to address this subject, including the Theory of Reasoned Action (TRA) (Ajzen and Fishbein, 1977), Technology Acceptance Model (TAM) (Davis, 1989), Theory of Planned Behaviour (TPB) (Ajzen, 1991), Unified Theory of Acceptance and Use of Technology (UTAUT) (Venkatesh et al., 2003), and its extension, UTAUT2 (Venkatesh et al., 2012; Arfi et al., 2021), which often serve as the foundation for technology adoption studies. One of the most commonly used theoretical frameworks is the UTAUT approach introduced by Venkatesh et al. (2003).

Beyond the original TAM, which focuses on Perceived Usefulness (PU) and Perceived Ease of Use (PEOU) (Davis, 1989), subsequent extensions such as TAM2 (Venkatesh and Davis, 2000) and TAM3 (Venkatesh and Bala, 2008) have expanded our understanding of technology acceptance. TAM2 introduced Social Influence (e.g. subjective norm, image) and cognitive instrumental processes (e.g. task relevance, output quality) as new determinants of Perceived Usefulness, emphasizing that technology use is shaped not only by individual beliefs but also by contextual and normative pressures. TAM3, in turn, integrates the determinants of Perceived Ease of Use, including computer self-efficacy, anxiety, enjoyment, and perceptions of external control. This version recognizes that ease of use results from both cognitive and emotional evaluations. Building on these developments, the augmented TAM integrated additional variables, such as Trust, Perceived Risk, System Quality, and Information Quality (Binyamin and Hoque, 2020), thereby improving its predictive validity in a consumer-oriented context such as IoT adoption. These successive refinements are conceptually connected to UTAUT and UTAUT2, which serve as meta-frameworks synthesizing the main determinants from TAM2 and TAM3. Specifically, UTAUT’s Performance Expectancy (PE) and Effort Expectancy (EE) correspond to TAM’s Perceived Usefulness and Perceived Ease of Use, while Social Influence (SI) and Facilitating Conditions (FC) capture external and contextual dimensions (Venkatesh et al., 2003). The latter extension, UTAUT2, adds Hedonic Motivation, Price Value, and Habit to better account for consumer contexts (Venkatesh et al., 2012). Hence, the trajectory from TAM through TAM2/TAM3 to UTAUT/UTAUT2 illustrates a gradual theoretical expansion from individual-level cognitive evaluation to the inclusion of social, emotional, and habitual factors relevant to technology use.

In this study, we integrate concepts from TAM and UTAUT by introducing Attitudes as a mediating construct between UTAUT predictors and Behavioural Intention. This integration allows examination of both direct and indirect effects of UTAUT constructs on consumers’ Behavioural Intention through Attitudes toward IoT. The conceptual novelty of this approach lies in bridging the attitudinal focus of TAM with the structural comprehensiveness of UTAUT, providing a more holistic understanding of technology acceptance. While UTAUT assumes that its determinants directly predict Behavioural Intention (Venkatesh et al., 2003), TAM emphasizes the mediating role of Attitudes as proximal drivers of intention (Davis, 1989; Fishbein and Ajzen, 1975).

By combining both frameworks, we propose a hybrid model that captures cognitive, social, and affective determinants of IoT adoption. This approach extends current knowledge by demonstrating that attitudinal processes can mediate the influence of core UTAUT constructs – particularly in contexts characterized by limited consumer experience and high uncertainty, such as IoT-based food technologies. This integration is consistent with recent meta-analytical findings (Binyamin and Hoque, 2020; Ladeira et al., 2025) that highlight the importance of attitudes in early-stage technology diffusion.

The article aims to comprehensively explore the factors impacting consumers' willingness to adopt IoT in food practices using the modified UTAUT model. Drawing on our research findings, we address the knowledge gap in consumer responses to IoT in the context of food consumption and provide insights for future studies.

The UTAUT model posits that Behavioural Intention and actual use of technology are influenced by four key determinants (Marikyan et al., 2023; Wang et al., 2021): Social Influence (SI), Performance Expectancy (PE), Facilitating Conditions (FC), and Effort Expectancy (EE) (Venkatesh et al., 2003). UTAUT synthesizes elements from eight well-established models used to investigate technology acceptance, including TRA (Theory of Reasoned Action), TAM (Technology Acceptance Model), MT (Motivational Model), TPB (Theory of Planned Behaviour), combined TAM and TPB, MPCU (Model of PC Utilization), IDT (Innovation Diffusion Theory), and SCT (Social Cognitive Theory) (Ramantoko et al., 2016).

The longitudinal research test found that the UTAUT model can explain 70% of the variance in Behavioural Intention, which is much better than any of the original eight models (Wang et al., 2021). Subsequently, the UTAUT model was expanded and refined into UTAUT2 by introducing three new constructs: Hedonic Motivation, Price Value, and Habit to highlight acceptance in the consumer context (Venkatesh et al., 2012). According to Wang et al. (2021) “compared with the UTAUT, the extension proposed in the extended UTAUT2 has substantially improved behavioural intention and differences in interpretation in technical use”.

However, as argued by Ramantoko et al. (2016) and Yang et al. (2024), during the initial stages of IoT solutions entering the food market, consumer experience is typically low. Consequently, determining the significance of habit or hedonic motivation may prove challenging. Similarly, assessing the value of price from the customer's perspective is complex; smartphone applications are often free, while the cost of IoT-equipped devices such as smart refrigerators, stoves, scales, and wearables may be perceived as high by some consumers. Therefore, in our study, we adopted a theoretical framework based on the UTAUT and TAM model and included consumers' perceived trust and security towards a relatively unknown technology as factors that could significantly influence their adoption considerations. This section describes the modified UTAUT constructs and theoretical model.

Performance Expectancy of IoT in the area of food consumption (PE) refers to the extent to which an individual believes that using IoT technology will help in food purchases, save time in meal preparation, or reduce food waste. IoT data streams from smart devices and household appliances support AI-powered features, such as waste-reduction alerts and personalised dietary suggestions. Consumers are typically more motivated to adopt IoT if they perceive it as beneficial and practical in their daily routines. This description of PE aligns closely with the perceived usefulness construct found in TAM, TAM2, TAM3, and the augmented TAM (Binyamin and Hoque, 2020). Previous research indicates that consumers' perceptions of the benefits and usefulness of technologically innovative solutions significantly influence their intention to use them. This trend is evident in various contexts, such as mobile payments (Slade et al., 2015), including restaurant payments (Khalilzadeh et al., 2017), augmented reality applications in retail (Jajić et al., 2021), drone food delivery (Choe et al., 2021; Leong and Koay, 2023), blockchain implementation in ancient wheat supply chains (Bandinelli et al., 2023), and smartphone adoption among older adults (Yang et al., 2022). To verify whether similar relationships also apply to IoT solutions, we formulated the following hypotheses.

Effort Expectancy of IoT in the area of food consumption (EE) refers to the perceived ease or difficulty of using the technology. It indicates the degree to which individuals believe that using a specific technology will require effort. EE is a crucial factor in the acceptance of innovative technologies. Using IoT data, AI can help people automatically create shopping lists and meal plans, so routine decisions can be made without manual effort. The simplicity and intuitive operation of IoT devices play a significant role in encouraging the adoption of innovative solutions. Previous studies have shown a significant positive impact of EE on BI in various contexts, including m-learning (Chao, 2019), drone food delivery (Choe et al., 2021; Leong and Koay, 2023), blockchain (Bandinelli et al., 2023) and augmented reality application acceptance (Jajić et al., 2021). Moreover, the intention to use website services can be predicted based on PE and EE (Jaradat and Banikhaled, 2013). Based on the above, we hypothesized the following.

The Social Influence of IoT (SI) refers to the extent to which a consumer perceives the influence of “significant others” in their environment on the adoption of a new technology. Derived from the Theory of Reasoned Action and the Theory of Planned Behaviour (Ajzen, 1991), SI recognizes that food purchases and eating practices often involve socialization processes and the impact of social pressure. Subjective norms serve as a tool of social impact, either supporting or inhibiting specific behaviours of an individual. Research by Okumus et al. (2018) demonstrated that Social Influence significantly affects the intention to use diet apps, and Khalilzadeh et al. (2017) found similar effects for mobile payments in restaurants. Additionally, findings by Limsarun et al. (2021) suggest that Social Influence is the most significant factor influencing the acceptance of food ordering applications and the intention to use conversational robots (De Andrés-Sánchez and Gené-Albesa, 2023). Perceived Social Influence also has a significant and positive impact on behavioural intentions to use digital wallets (Khan and Abideen, 2023). Positive Social Influence on acceptance intention has been observed in studies on mobile payments, autonomous cars, mobile health services, and insects as food (Alam et al., 2018; Berger et al., 2019; Leicht et al., 2018; Slade et al., 2015). Therefore, we formulated the following hypotheses.

Researchers often augment the UTAUT model with various constructs, such as Perceived Risk (Martins et al., 2014), Brand Trust (Rehman et al., 2022), or Trust in technology (Binyamin and Hoque, 2020). One challenge in implementing IoT solutions is the data transmitted by applications installed on consumers' smartphones and other sensor-equipped devices. Therefore, in our study, we included consumer trust toward IoT as a criterion.

Trust in IoT (TR) refers to consumers’ belief that technology is reliable, efficient, and safe, and that it protects transmitted data. When users believe that technology will deliver the desired results, it strengthens their behavioural intention. Trust has been identified as a crucial factor in the acceptance of healthcare technology. TR is a prerequisite for users’ acceptance of electronic health services, such as e-Health/mHealth (Alam et al., 2018). This aspect is also considered the most significant factor for predicting the adoption of mobile health. However, opposing conclusions have been drawn regarding wearable health monitoring technology (Binyamin and Hoque, 2020). It is assumed that food consumers with higher levels of trust will be more willing to adopt IoT, which is why this variable was included in our research. It is not present in the original UTAUT model; however, it was added in its newer versions (Arfi et al., 2021). We assume that TR can be seen as a factor preceding attitude and behavioural intention:

In the original UTAUT model of Venkatesh et al. (2003), one more variable was specified: facilitating conditions (FC). FC refers to the extent to which a user believes that supportive conditions exist for using technology. Consumers' belief that they have the necessary knowledge and resources translates into a willingness to use the Internet of Things. Conversely, the perception of inadequate conditions will likely limit the use of IoT. We have omitted this construct in our research because it is linked to the purchase in the UTAUT model. In turn, we included Attitudes and Behavioural Intention as intended by (Davis et al., 1989).

Attitudes (ATT). The main assumption of the TAM model is that behavioural intention is shaped by attitudes toward technology use (Fishbein and Ajzen, 1975). The results of studies of Davis (1985) and Zahid et al. (2010) show that consumers' attitudes toward adopting a new technology are crucial to its adoption. It should be noted that consumer attitudes toward new technologies such as IoT can be dynamic and evaluative (Fazio, 2007). The valence of IoT attributes is difficult for consumers to assess, as they lack adequate knowledge or experience with technology. Thus, the process of attitude formation involves constructing evaluations of the technology being evaluated. These evaluations can result from the impact of objective, fragmentary information, from a certain emotional reaction triggered by the technology's attributes, and from signals/opinions from the consumer's environment.

The process of formulating consumer attitudes toward IoT occurs in a specific context, where the influence of cognitive resources (versus affective resources) may be small, given the increasing importance of an individual's interaction with the social context to which they belong (Rigopoulou et al., 2015). We assume that AT is a mediating variable influencing behavioural intention.

Behavioural intention (BI). In our study, we embraced a behavioural approach (Khalilzadeh et al., 2017), viewing intention (BI) as a driver of consumer behaviour. It reflects the extent to which a consumer intends to carry out specific actions in the future. By assessing Behavioural Intention, we can forecast the consumer's actual behaviour towards technology. Intention, a person's inclination to use technology, predicts real-world behaviour. Based on the Theory of Planned Behaviour (TPB) (Ajzen, 1991), Behavioural Intention is preceded by Attitude, subjective norms, and behavioural control. The Behavioural Intention has abundant empirical verification of consumer acceptance of modern technologies in the food market, including the use of food delivery apps (Limsarun et al., 2021), the purchase of fast food with digital coupons (Akram et al., 2020), and the use of self-service kiosks in food service (Jeon et al., 2020). We assume that, for early-stage products, the constructs suggested by the UTAUT model will not form separate entities but will be similar to one another and constitute sub-dimensions of the overall Attitudes construct. We, therefore, hypothesized the following:

To test our hypotheses, we developed a theoretical model (see Figure 1).

It allows us to examine how Performance Expectancy, Effort Expectancy, Social Influence, and Trust directly and indirectly influence Behavioural Intention through Attitudes. The pathways are based on the theoretical assumptions described above.

The data was collected by the research company Biostat in April 2022. 600 panelists completed the questionnaire using a web application or a mobile app, available on Android and iOS. The study was conducted in agreement with the Declaration of Helsinki. The quota sampling was used, with the following selection criteria: gender, age, and place of residence (95% confidence interval). Respondents were selected non-randomly within quotas, making this a non-probability design. The sample exhibited a nearly equal gender distribution, with 49.7% of subjects identifying as female and 50.3% as male. The respondents were aged 18–59, and their age distribution was representative of Polish internet users. Most had secondary education (47.0%) or higher (38.7%), and 60.0% lived in towns or cities (compared to 40.0% in villages). The majority rated their financial situation as at least average, with 48.0% describing it as average and 43.0% as good or very good. Detailed characteristics of the respondents are presented in Table A1 in the Appendix.

Recognizing that the term Internet of Things might be unclear to respondents and that they might be unfamiliar with its uses, at the start of the survey, respondents were introduced to the IoT definition and examples of how IoT can be applied in the process of buying food, preparing meals, and customizing diets. This introduction is presented below:

All constructs of the research model (Figure 1) were estimated by using multiple items in reflective measurement models (Sarstedt et al., 2016) adopted from previous studies: the IoT in general (Lee and Shin, 2019; Mącik, 2017), adoption intention of IoT-based wearables for the health care of older adults (Sivathanu, 2018), adoption of solar energy panels (Claudy et al., 2013), behavioural reasoning theory (BRT) (Westaby et al., 2010), consumers’ adoption of mobile shopping (Gupta and Arora, 2017), new product adoption (Wang et al., 2008). Individual items were modified to reflect the use of IoT in food consumption. Within the individual constructs, items from different scales were used. For example, for the attitude construct, items were drawn from three different scales proposed by (Sivathanu, 2018; Tsourela and Nerantzaki, 2020; Venkatesh et al., 2012). Items selection was made by 3 expert judges (the basic criteria for selection were their validity, non-redundancy, and their fit with the Polish language). In one case, related to Social Influence, we included an item not present in other scales, such as “Famous and popular people in the media use the Internet of Things.” All items were measured on a 5-point Likert scale ranging from 1 – “I completely disagree” to 5 – “I fully agree”. The full set of items, with corresponding sources, is presented in Table A2 in the appendix. Descriptive statistics for the constructs are presented in Table A3. The Committee on Research Ethics accepted the study at the Poznań University of Economics and Business.

The selection of the data analysis method was guided by the sample size, the model's complexity and predictive nature, and the requirement to apply bootstrapping for the mediation analysis. These criteria were largely met by applying PLS-SEM, which enables the estimation of complex and prediction-oriented models (Hair et al., 2019) and is well-suited for conducting advanced mediation analyses (Nitzl et al., 2016; Sarstedt et al., 2020). In addition, PLS-SEM is often used in the field of innovation acceptance (e.g. Faiz et al., 2024; Tu et al., 2021), thereby further increasing its relevance. Due to the use of reflective constructs, the PLSc-SEM method was applied (Dijkstra and Henseler, 2015). Additionally, the inclusion of map analysis led to the adoption of an application perspective. SEM identifies statistical relationships, but cIPMA extends this approach by providing practical guidance to decision-makers (Hauff et al., 2024; Sarstedt et al., 2024). The data were screened for inattentive respondents using the Longstrings method (Curran, 2016), reducing the sample to 544 subjects.

We carried out the analysis in two stages, assessing the measurement (or outer) model and structural (or inner) model (Anderson and Gerbing, 1988). The measurement model was tested using confirmatory composite analysis (Hair et al., 2020). For the data analysis, we used the procedure proposed by Hair and Alamer (2022), supplemented by suggestions from recent publications (Becker et al., 2022; Sarstedt et al., 2023).

The assessment of outer loadings revealed that several indicators failed to meet the recommended threshold of 0.7, as suggested by Hair and Alamer (2022). Among these, EE_2R loading was notably low, likely due to its reversed-scoring format. EE_2 R was removed from the outer model, and all other low outer-loading items were retained, as all constructs met the recommended thresholds for validity and reliability (Hair and Alamer, 2022).

In addition, we conducted a bootstrapping analysis (10,000 re-samples) to obtain confidence intervals for the assessed criteria. Assessment of the measurement model indicated that reliability, internal consistency reliability (Composite Reliability and rho_A), convergent validity, and discriminant validity are confirmed (see Table A4 appendix). In essence, all AVE values were above the 0.5 threshold, composite reliability values ranged from 0.850 to 0.944, and Cronbach's alpha values ranged from 0.881 to 0.946, and in both cases were inside the suggested range of 0.7–0.95 (Hair and Alamer, 2022; Henseler et al., 2015). A standard heterotrait-monotrait ratio (HTMT) test (without bootstrapping) showed that none of the values exceeded the threshold of 0.9 (Lowry and Gaskin, 2014) (see Table A5, appendix). An HTMT test (Henseler et al., 2015) with bootstrapping showed that none of the confidence intervals contained the value 1 (Table A4 appendix).

We performed the structural model assessment in the four stages suggested by Hair and Alamer (2022). In the first stage, we assessed the potential threat posed by collinearity. We used the value inflation factor (VIF) to accomplish this task. The VIF values ranged from 2.322 to 4.640, thus exceeding the threshold of 3 indicated by Hair and Alamer (2022), although it was below the more benign threshold of 5 indicated in earlier publications (Hair et al., 2017). The collinearity problem involved the two variables, Behavioural Intention and Attitude toward IoT. These dictate caution in interpretating the results, suggesting a potential problem with the model. To address the use of reflective variables on the same measurement scale (Podsakoff et al., 2003), we also conducted a test for common method bias. We applied the approach suggested by (Kock, 2017), which requires that in a full test for collinearity, all VIF values should remain below the threshold of 3.3. We performed an analysis using a random variable (a common latent factor) as the dependent variable. The results showed that none of the values exceeded the recommended threshold of 3.3 (see Table A6 in the appendix).

In a second step, we assessed the direct path coefficients using a bootstrapping procedure (10,000 re-samples, Bias-Corrected and Accelerated confidence interval estimation method, and two-tailed testing at the 0.05 level). We used f² to assess the importance of each predictor in explaining the endogenous constructs. Typical thresholds for interpreting f², which reflect the increase in explained variance after including a specific variable, are 0.02 (small effect), 0.15 (medium effect), and 0.35 (large effect) (Henseler et al., 2009). The results are presented in Table 1.

None of the control variables had a significant impact on BI (p > 0.05; see Table 1). Nevertheless, including these variables increased the model's robustness (Gudergan et al., 2025; Klarmann and Feurer, 2018), indicating that the examined relationships between constructs were not disturbed by age, financial situation, and education.

The results showed that three constructs had a direct effect on the ultimate outcome variable (BI). The first was ATT (β = 0.655, t = 7.635, p < 0.001), which was the strongest predictor with a large effect size (f² = 0.759). The second was SI (β = 0.349, t = 6.254, p < 0.001) with a large effect size (f² = 0.405) and PE (β = 0.123, t = 2.263, p < 0.05) with a small effect size (f² = 0.034). There results provide support for hypotheses: H3, H1a, and H1c.

Two remaining relationships between EE and BI (β = −0. 052, t = 0. 622, p > 0.05) as well as between TR and BI (β = −0.086, t = 1.354, p > 0.05) led to statistically insignificant results. Thus, hypotheses H1b and H1d have not been supported. Neither control variable reached the significance level (see Table 2).

When considering the mediating variable ATT, the results indicate that all four exogenous path coefficients of the constructs were statistically significant. These were TR (β = 0.411, t = 5.997, p < 0.001, with medium effect size (f² = 0.305), subsequently examined PE (β = 0.200, t = 2.300, p = 0.021), with small effect size (f² = 0.054), EE (β = 0.219, t = 2.466, p = 0.014), with small effect size (f² = 0.078), and SI (β = 0.155, t = 2.466, p = 0.014), with small effect size (f² = 0.056). These results support hypotheses: H2a-H2d.

We also analyzed the mediating effect of Attitudes IoT on Behavioural Intention, as the significance of the direct path does not always imply the significance of mediation (Zhao et al., 2010). The results of mediation are presented in Table 2. They indicate that ATT has significant mediating effects on the relationship among PE, EE, SI, TR, and BI, as none of the confidence intervals include 0, providing support for the H4a-H4d hypotheses.

We applied the procedure proposed by Gaskin et al. (2023) to estimate the size of indirect effects. The results indicated that the indirect effect sizes were small except for the relationship related to TR, in which case the indirect effect size was medium (see Table 2).

In the third step of the procedure proposed by (Hair and Alamer, 2022), we used the coefficients of determination to assess the structural model's explanatory power. The value of R² ranges from 0 to 1, but its interpretation is not clearly defined (Chicco et al., 2021; Helland, 1987). The model explained 77.6% of the variance in ATT and 86.9% of the variance in BI. It is worth noting that the R² values are context-dependent (Kopplin and Rausch, 2022). To compare them, data from other similar studies would need to be used to establish an appropriate interpretation. Given that there are no such directly comparable studies, the interpretation of R² should be treated with caution. In other studies in the area of IoT adoption that applied a similar methodological approach, the R² value for Behavioural Intentions (BI) in the context of implementing voice user interfaces was 0.64 (Song et al., 2022), while for the adoption of smart home technologies, the adjusted R² value reached 0.523 (Gøthesen et al., 2023). Considering that R² value depends on the number and strength of predictors (with four predictors included in the model), the R² values in our study can be considered substantial.

In the fourth step, we applied the PLS predict procedure (Guenther et al., 2023) to assess the model’s predictive power in new samples. The results (see Table A7, appendix) of the analysis showed that for most of the endogenous variable indicators, the prediction errors (in terms of RMSE or MAE) were lower than those of the naive linear regression model (LM). This indicated that the model had a medium out-of-sample predictive power (Hair and Alamer, 2022).

From a practical point of view, it is vital to identify the key variables for a successful introduction of IoT products or services in food consumption. To facilitate the interpretation of the results in the context of practical application, the recently developed cIPMA (Hauff et al., 2024; Sarstedt et al., 2024) was applied, presenting in a synthetic way the combined performance and importance of the predecessor latent variable of PE, EE, SI, TR, and ATT in achieving a high score on the Behavioral Intention construct. The first dimension of cIPMA is “importance”, which reflects the constructs’ total effects. The second dimension is the “performance”, which represents the mean value of their rescaled latent variable scores (ranging from 0 to 100). The third component, expressed in terms of the size of the circles, refers to the percentage of people who did not reach the set performance level of the target variable. cIPMA can be presented in the form of a table or a map (see Figure 2). To create cIPMA, we used the Excel template that was available on SmartPLS website (https://www.pls-sem.net/downloads/additional-useful-downloads/).

The bubble sizes indicate the percentage of cases that have not achieved the 75% level outcome of Behavioural Intention. Black dots denote constructs below the threshold of practical relevance.

As the distinction between low and high performance/importance is quite arbitrary (Hauff et al., 2024; Streukens et al., 2017), we set performance cut-off values at 75% (representing a high but non-extreme level of the construct). For importance, we followed Streukens et al. (2017) and set the importance at 0.3, corresponding to a medium effect size. Cut-off values enabled us to divide the map into four areas whose standard interpretation was formulated by Streukens et al. (2017).

The cIPMA map (see Figure 2) indicates a relatively similar level of performance and a large variation concerning importance. None of the predecessor variables were in the upper parts of the map, nor on the left (interpreted as possible overkill), nor on the right (keep up the good work). All variables preceding BI were placed at the bottom of the map.

As only two variables reached the level of practical relevance, the interpretation will be focused on them. Although they exhibit a similar level of performance, their interpretation differs significantly due to varying levels of importance. While PE remains a key factor, EE can no longer be considered as such. EE is situated in the “low importance” area, indicating that efforts to promote IoT adoption should not prioritize this factor.

Ultimately, therefore, of the relatively large number of predecessor variables, the practical focus should be on just one, PE. This construct is in the “Concentrate here” area, which emphasizes not only the importance, but also its development potential.

The highest level of importance was observed for the latent variable Attitude. However, this variable is practically irrelevant because the required 75% level has already been reached by all subjects. The next two variables (SI and TR) were also practically irrelevant.

Based on the modified UTAUT model by Venkatesh et al. (2003), this study analyzed the factors shaping Attitudes and Behavioural Intention regarding IoT in food consumption. The standard UTAUT model was extended by incorporating Trust and Attitudes toward IoT as mediating variables. Our findings lead to two key conclusions: the critical role of PE in shaping Behavioural Intention, and the lack of justification for including Attitudes in the UTAUT model.

When applying cPLS-SEM, a particularly notable effect was observed in the relationship between ATT and BI. cPLS-SEM revealed a strong relationship between these variables, consistent with prior studies, e.g. (Borusiak et al., 2021; Chatterjee et al., 2021). However, this strong relationship can also be due to automatic decision-making or, in the case of research, automatic question answering, which can be linked to dual-process theories (Grayot, 2020; Osman, 2004) or to some redundancy between these two variables.

An additional key finding relates to Performance Expectancy (PE). PE proved to be a significant factor influencing BI. The cIPMA analysis indicated that PE is in a critical zone, highlighting the need for further efforts to improve consumers’ perceptions of IoT. Despite the positive impact of PE on BI, the performance level of 75% has not yet been achieved. External information sources, such as advertisements (Hoeffler and Keller, 2002; Parris and Guzmán, 2023) and social media (Leung et al., 2022; Lutfi et al., 2023), play a significant role in shaping PE. Therefore, continued investment in education and promotion is necessary to enhance consumer perceptions.

Another conclusion concerns the impact of social influence on BI. The results of cPLS-SEM indicated a strong relationship between these two variables, whereas cIMPA showed low importance of SI. Thus, the results didn’t show a clear interpretation of the role of SI. Some research, for instance, Limsarun et al. (2021) identified Social Influence as the most critical determinant in the adoption of food ordering applications. Broadly speaking, it can be inferred that the social pressure felt by consumers encourages or compels them to take certain actions, thereby significantly influencing their Behavioural Intention. Such claims may be difficult to generalize, as results may be context-dependent. For example, IoT in the food consumption domain may not be perceived as socially consumed products, such as fashion products (Kamal et al., 2013) or luxury products (Iyer et al., 2022), and therefore, the impact of SI on purchasing decisions may be small.

Considering our research, Trust was not a significant variable. The low importance of this factor is surprising, as many publications emphasize its high role in the acceptance of novelty (Arfi et al., 2021; Patil et al., 2020). In light of the results, it can be concluded that Trust is either not important or that consumers' beliefs have not yet matured to the point of considering it an important dimension. In the latter case, this would argue for an early stage of belief development in this sphere.

The results of the cPLS-SEM study indicate that Effort Expectancy (EE) did not have a significant direct impact on Behavioural Intention. However, the cIPMA analysis indicated that this variable was of low importance. Previous studies with varying degrees of success documented the effect of EE on Behavioural Intention. There are several examples indicating that there is no relationship between EE and BI, e.g. (Abushakra et al., 2022; Hossain et al., 2019). Unambiguously identifying the reasons this variable is significant in some conditions but not in others is difficult. One potential solution to this issue is the aforementioned stage of attitude development. At this preliminary stage, respondents have yet to articulate or quantify the ease or complexity of utilising the Internet of Things (IoT), or whether it necessitates specific skills or expertise.

Although Trust and Effort Expectancy did not affect BI, their lack of influence is nonetheless a valuable insight into the topic under investigation. Contrary to our findings, numerous studies have identified such relationships, as evidenced by research in the domain of organic cosmetics (Munerah et al., 2021), IoT in eHealth (Arfi et al., 2021), IoT in smart home context (Aldossari and Sidorova, 2020), and IoT-enabled devices (Chatterjee, 2022). The documentation of studies in which such relationships are absent is relatively rare. This situation could be attributed to publication bias, i.e. the tendency to underreport studies with nonsignificant results (Brodeur et al., 2022; Franco et al., 2014). Nonetheless, acknowledging and publishing findings of non-significant impacts is crucial for a comprehensive understanding of the phenomena in question.

Our research also indicated a lack of justification for introducing Attitudes as a mediating variable in the UTAUT model, as indicated by cIMPA. On the one hand, Attitudes proved to be an important mediator in the cPLS-SEM analysis. However, cIPMA showed this variable to be of low importance. The direct impact of Attitudes towards IoT on Behavioural Intention is well-established in the literature, as highlighted by numerous studies (e.g. George et al., 2021; Van Der Heijden et al., 2003). However, it is worth noting that these studies often made different theoretical assumptions and were most often based on the TAM model (Costa et al., 2021; Madias et al., 2023). What distinguishes our study is that the effects of all four constructs on Attitudes are statistically significant, despite the absence of such effects (apart from SI and TR) on Behavioural Intention.

This emphasizes the potentially important role of Attitudes in the acceptance of new technologies, supporting the notion that consumer attitudes during the initial acceptance phase are generally broad and somewhat superficial (Tang et al., 2003). Over time, as consumers gain more information and personal experience, these attitudes develop into more complex and multivariate phenomena (Ajzen, 2001; Barth, 2016; Kraus, 1990). Therefore, evaluating constructs at the early stage of product acceptance may lead to overinterpretation of the data, potentially resulting in illusory correlations between variables (Fiedler et al., 2022), which can complicate the replication of research findings (Camerer et al., 2018).

Conversely, moderate and weak Behavioural Intention appears to be loosely dependent on Attitudes, with the correlation identified in the cPLS-SEM analysis attributable to acquiescence bias (Sakshaug, 2021; Sakshaug and Kreuter, 2012) rather than a meaningful correlation. In addition, there was a collinearity problem between ATT and BI. The arguments against ATT are therefore stronger than those in favor of including this construct in the UTAUT model.

Our research showed no significant impact of the control variables. These were age, financial situation, and education. As observed in other studies, these results indicate that psychological and social factors outweigh demographic and economic factors in IoT adoption (Dwivedi et al., 2019; Qi et al., 2021; Schroeder et al., 2023).

The findings underline the potential of IoT technologies to foster sustainability-oriented food consumption. In line with previous studies, IoT enhances transparency and traceability across the agri-food chain, enabling consumers to make informed choices about environmentally responsible, and low-carbon products (Kamble et al., 2019; Wolfert et al., 2017; Caferra et al., 2023; Yamoah et al., 2022). By improving access to real-time data on product origin and resource use, IoT supports awareness of sustainability benefits and encourages behavioural shifts toward responsible food practices. These insights complement earlier research on digital traceability and eco-innovation in food systems (Bandinelli et al., 2023; Jeon et al., 2020), reinforcing the view that technology acceptance can drive sustainable consumption. Summarizing, IoT devices supported by AI solutions for intelligent decision-making have the potential to influence consumer behaviour, including how food is purchased, stored, prepared, consumed, and disposed of (Dakhia et al., 2025).

In this study, we explored the factors shaping attitudes and behavioural intention regarding IoT in food consumption. The standard UTAUT model was extended by incorporating Trust and Attitudes toward IoT as mediating variables in the adoption of food-related IoT, offering evidence from an under-explored context. The findings, based on both cPLS-SEM and cIPMA analyses, reveal that performance expectancy (PE) is the most influential determinant of Behavioural Intention, while Social Influence plays a secondary role. Contrary to expectations, Trust and Effort Expectancy did not directly predict Behavioural Intention, although both significantly shaped attitudes toward IoT. These results suggest that early-stage adoption is primarily driven by perceived utility rather than ease of use or Trust considerations. The strong mediating role of Attitudes underscores the complexity of consumer evaluations during initial exposure to novel technologies, supporting the notion that Attitudes are broad and somewhat superficial in early adoption phases (Ajzen, 2001; Dwivedi et al., 2019). Overall, IoT adoption in food contexts hinges on perceived performance benefits and normative cues, offering actionable insights for accelerating digital transformation and sustainability in the food sector.

Our research provides policymakers, marketers, and developers with valuable guidance on promoting the adoption of IoT in food consumption. AI, driven by data from IoT-connected appliances and sensors, offers new opportunities for marketers to enhance the consumer experience (Dabija and Frau, 2025). By addressing the identified factors and leveraging Social Influence and targeted interventions, stakeholders can foster a favorable environment for the widespread acceptance and adoption of IoT-enabled solutions that support sustainable food consumption in this domain. In the context of consumer communication, it is necessary to underscore the potential of these technologies to improve quality of life, facilitate sustainable consumption decisions, and save time and financial resources.

The findings of this study offer valuable insights into IoT acceptance and adoption in food consumption and highlight how IoT–AI integration can be applied in consumer contexts. We contribute to refining the Unified Theory of Acceptance and Use of Technology (UTAUT) model by integrating Trust and Attitudes towards IoT as mediating variables, thereby extending understanding of the factors influencing Behavioural Intention in this domain. We show that Attitudes act as mediators even when their practical role is constrained during early stages of adoption. The non-significance of Trust and Effort Expectancy challenges prior IoT research, suggesting that perceived utility takes precedence when user experience with the technology is minimal.

The identified factors influencing Behavioural Intention, particularly Performance Expectancy (PE), underscore the need to design interventions and promotional strategies that target them (Philippi et al., 2021). Given the critical role of PE, efforts should focus on enhancing consumers' perceptions of the performance benefits of IoT applications in food consumption. Potential strategies include targeted advertising campaigns, product demonstrations, and leveraging social media platforms to highlight the advantages of IoT-enabled solutions.

The findings provide practitioners with actionable guidance: enhancing perceived performance value, simplifying the user experience, and strengthening Trust through transparent data practices can meaningfully increase consumer acceptance of IoT solutions that support sustainable food consumption. Marketing strategies should prioritize communicating specific benefits, such as time efficiency, waste minimization, and tailored nutritional solutions, to enhance performance expectations. Social influence may continue to play a significant role, particularly through collaborations with influencers and community-driven campaigns (Okumus et al., 2018; Limsarun et al., 2021). While Trust did not emerge as a direct determinant, transparent data governance practices and recognized certifications are likely to gain prominence as consumer awareness increases. Policymakers should prioritise protecting consumers' interests while also ensuring consumer education in this domain. Furthermore, retailers and IoT solution providers are encouraged to incorporate AI-enabled personalization and sustainability-oriented features to align their offerings with evolving consumer expectations and accelerate digital transformation in food systems (Dakhia et al., 2025; Frau et al., 2022).

IoT adoption in food consumption can accelerate sustainable behaviours by reducing food waste and carbon footprints through data-driven decision-making. Policymakers should support digital literacy programs and incentivize IoT-enabled solutions that promote health and sustainability. The integration of IoT and AI in food systems aligns with global sustainability goals, fostering responsible consumption patterns and reducing resource inefficiencies (Gbashi and Njobeh, 2024).

The present research has several limitations. First, the use of a quota sampling method may limit the extent to which the findings can be generalized. Second, the geographical scope is limited to Poland, suggesting that future studies should conduct international comparisons to examine whether the observed relationships hold across different cultural and market contexts. Third, the application of IoT to food consumption is not yet widespread in Poland, so despite explanations of what IoT is and how it can be helpful, respondents may have had difficulty answering the questions; without prior experience of this technology, it is challenging to imagine its usefulness in everyday life or the problems it may cause. It can be assumed that the Effort Expectancy questions were particularly problematic for respondents. Therefore, it is advisable to apply the UTAUT model primarily to technologies with which consumers have at least some experience, rather than to entirely new solutions. Furthermore, integrating the Internet of Things (IoT) into food consumption requires careful consideration alongside artificial intelligence (AI), given the potential interplay among Trust, Social Influence, and Attitudes shaped by AI-generated guidance derived from IoT data. Finally, the proposed model focuses on consumers’ intention to use IoT in food practices. At the same time, any translations of IoT data into conclusions using AI/ML are not based on empirical analysis, but on analysis of literature and previous research in this area.

Future research should pursue cross-cultural validation to examine whether social influence and Trust gain prominence in collectivist versus individualist societies (Venkatesh and Zhang, 2010; Giacalone and Jaeger, 2023). Longitudinal studies are needed to track changes in Attitudes and Behavioural Intentions as consumers gain experience with IoT, and to assess whether Trust and Effort Expectancy become more influential over time. Further investigation into the interplay between IoT and AI is warranted, particularly regarding consumer Trust in AI-generated recommendations for personalized nutrition (Dakhia et al., 2025). Comparative studies across sectors such as energy and water could test the generalizability of findings to other domains of sustainable consumption (Caferra et al., 2023). Finally, experimental interventions should evaluate the effectiveness of targeted communication strategies – such as sustainability framing and influencer campaigns – in enhancing performance expectancy and adoption rates (Hoeffler and Keller, 2002; Leung et al., 2022).

The supplementary material for this article can be found online.