Dissertation – Artificial Intelligence Adoption in Australian Small and Medium Enterprises

Abstract

Generative artificial intelligence (AI) has moved rapidly from novelty to mainstream business tool, yet adoption among small and medium enterprises (SMEs) remains uneven. This dissertation examines the factors shaping generative AI adoption among Australian SMEs, integrating the technology-organisation-environment (TOE) framework with the Unified Theory of Acceptance and Use of Technology (UTAUT). A convergent mixed methods design combined a survey of 320 owner-managers with 14 semi-structured interviews. Structural modelling showed that performance expectancy, relative advantage and effort expectancy were the strongest positive predictors of adoption intention, while data governance concern and perceived cost exerted significant negative effects; the model explained 61 per cent of the variance in intention. Interviews surfaced six themes centred on skills gaps, uncertain returns, trust in output accuracy, and privacy obligations under the Privacy Act 1988. The findings indicate that targeted capability building and clearer governance guidance would materially lift adoption and expected productivity across the Australian SME sector.

Introduction

Small and medium enterprises are central to the Australian economy, accounting for the overwhelming majority of actively trading businesses and a substantial share of private-sector employment (Australian Small Business and Family Enterprise Ombudsman [ASBFEO] 2023). Their capacity to adopt productivity-enhancing technology therefore has consequences well beyond the individual firm. Survey evidence indicates that while internet access is near-universal among Australian businesses, the use of more advanced digital capabilities remains uneven and is consistently lower among smaller firms (Australian Bureau of Statistics [ABS] 2023). The rapid public release of generative AI tools has sharpened this divide, raising the question of whether resource-constrained SMEs can capture the efficiency gains that larger organisations are already pursuing.

The Productivity Commission (2024) argues that the economy-wide benefit of AI will depend heavily on diffusion across smaller firms rather than adoption by a handful of large ones. Yet acceptance research has concentrated on individual users or large enterprises, leaving the firm-level dynamics of SME adoption under-examined, particularly for a technology as novel and fast-moving as generative AI. This study addresses that gap in the Australian setting. Its aim is to identify the factors that drive and constrain generative AI adoption among Australian SMEs and to assess how owner-managers perceive the resulting productivity outcomes. Three research questions guided the study:

1. Which technological, organisational and environmental factors influence generative AI adoption intention among Australian SMEs?
2. How do performance expectancy and effort expectancy shape owner-managers’ intention to adopt generative AI?
3. What barriers constrain adoption, and how do owner-managers perceive the likely productivity outcomes?

Literature Review

Firm-level adoption and the TOE framework

Firm-level technology adoption has long been theorised through the TOE framework, which locates adoption decisions in the interplay of technological, organisational and environmental conditions (Tornatzky & Fleischer 1990). The framework offers strong explanatory breadth across technologies and settings, but it is frequently criticised for under-specifying the psychological mechanisms that convert context into a decision (Oliveira & Martins 2011). For SMEs, the organisational dimension is distinctive because the owner-manager is typically the sole decision-maker, so individual disposition and firm resource constraints are tightly coupled rather than separable.

Acceptance beliefs and UTAUT

UTAUT addresses the mechanism gap by positing performance expectancy, effort expectancy, social influence and facilitating conditions as the proximal determinants of technology use, a synthesis later refined for consumer contexts as UTAUT2 (Venkatesh et al. 2003; Venkatesh, Thong & Xu 2012). Performance expectancy, the belief that a technology will improve task outcomes, is consistently the dominant predictor, while effort expectancy captures the ease with which a busy owner-manager can put a tool to work. Integrating UTAUT with the TOE framework allows contextual antecedents to be linked to belief-based mechanisms, a combination increasingly favoured in enterprise adoption research.

Generative AI in the Australian context

Three technological attributes recur in accounts of generative AI adoption: relative advantage, the trustworthiness of generated output, and data governance risk. National analysis suggests Australian firms see clear efficiency potential in AI but remain wary of accuracy and confidentiality (CSIRO 2023). Environmental pressures include competitive intensity and regulatory uncertainty; the latter is shaped in Australia by principles-based expectations under Australia’s AI Ethics Framework (Department of Industry, Science and Resources 2019) and by privacy obligations under the Privacy Act 1988, which govern how client information may be disclosed to third-party systems (Office of the Australian Information Commissioner [OAIC] 2023). Synthesising these strands, this study models adoption intention as jointly determined by TOE contexts and UTAUT beliefs, as summarised in the conceptual model presented in the following section.

Methodology

Research design

This study adopted a convergent mixed methods design, in which the quantitative and qualitative strands were collected in parallel and integrated at interpretation (Creswell & Plano Clark 2018). The design suits an emerging phenomenon in which measurable patterns benefit from explanatory depth that a survey alone cannot provide. Figure 1 illustrates the conceptual model tested in the quantitative strand and elaborated in the qualitative strand.

Participants and measures

The quantitative strand surveyed 320 Australian SME owner-managers, with SMEs defined per the standard convention of firms employing fewer than 200 people (ABS 2023). Respondents were recruited to vary by state and industry division to improve the generalisability of the sample (Bryman 2016), spanning retail, professional services, construction, manufacturing and hospitality. Constructs were measured with multi-item seven-point Likert scales adapted from validated instruments (Venkatesh et al. 2003), and the structural model was estimated using partial least squares structural equation modelling (PLS-SEM), which is appropriate for prediction-oriented studies with a mix of reflective and formative constructs (Hair et al. 2019). Convergent and discriminant validity were assessed through composite reliability, average variance extracted and the Fornell-Larcker criterion (Fornell & Larcker 1981).

Qualitative strand and ethics

The qualitative strand comprised 14 semi-structured interviews with owner-managers purposively selected to vary by adoption stage and sector. Interviews of 35 to 50 minutes were transcribed and analysed using reflexive thematic analysis, following the familiarisation, coding, theme development and review sequence (Braun & Clarke 2006). The study was approved by the administering university’s human research ethics committee, participation was voluntary and based on informed consent, and data were de-identified and stored consistent with the Australian Privacy Principles (OAIC 2023).

Findings

Structural model

Table 1 reports the structural model. Performance expectancy was the strongest predictor of adoption intention, followed by relative advantage and effort expectancy, with owner-manager innovativeness and competitive pressure significant but weaker. Two barriers were significant and negative: data governance concern and perceived cost. Regulatory uncertainty did not reach significance. Together the predictors explained 61 per cent of the variance in adoption intention.

Table 1: Standardised path estimates for predictors of generative AI adoption intention (n = 320)

Note. Standardised path coefficients (β) from PLS-SEM predicting adoption intention; n = 320; t and two-tailed p derived from bootstrapping. The model explained 61 per cent of the variance in adoption intention (R² = .61). Paths significant at p < .05 are marked supported.

Interview themes

The qualitative strand elaborated these patterns. Six themes were constructed from the interview data, summarised in Table 2. Owner-managers consistently linked expected productivity gains to routine drafting, summarising and customer-response tasks, yet tempered this optimism with concern about output accuracy and the effort of verification. Skills gaps and uncertainty about return on investment were the most frequently raised constraints, while privacy obligations were a recurring reason for caution when client data was involved. As one professional-services owner put it, “the time saving is obvious, but I cannot put a client’s file into something I do not control” (Participant 9).

Table 2: Themes from reflexive thematic analysis of owner-manager interviews (n = 14)

Discussion

The results support an integrated TOE-UTAUT account of generative AI adoption among Australian SMEs. Consistent with UTAUT, belief-based mechanisms dominated: performance expectancy and effort expectancy together outweighed environmental pressure, suggesting that owner-managers adopt when they can see concrete task-level benefit rather than in response to competitive signalling alone (Venkatesh et al. 2003). The prominence of relative advantage echoes the observation that efficiency is the leading motivation for Australian firms considering AI (CSIRO 2023), and it aligns with the interview theme in which productivity and time saving was the most widely shared expectation.

The negative effect of data governance concern is notable and corresponds to the qualitative finding that obligations under the Privacy Act 1988 shape how far owner-managers are willing to expose client information to third-party models (OAIC 2023). This points to a governance-readiness barrier that generic capability programs rarely address. The non-significant effect of regulatory uncertainty, despite frequent mention in interviews, suggests that principles-based guidance such as Australia’s AI Ethics Framework is not yet salient enough at the point of decision to deter adoption, although neither does it appear to reassure owner-managers (Department of Industry, Science and Resources 2019). The gap between what is voiced in interviews and what predicts intention is itself instructive: regulatory anxiety is expressed as a general unease rather than operationalised into the specific data-handling judgements that actually move the decision.

For policy and practice, the findings imply that diffusion across the SME sector, which the Productivity Commission (2024) identifies as pivotal to national productivity, depends less on awareness than on lowering effort and governance costs. The steep contribution of effort expectancy, mirrored by the skills-gap theme, indicates that practical, low-cost capability support would address one of the two binding constraints, while sector-specific templates for privacy-compliant data handling would address the other. Both would reinforce the productivity expectations that already motivate owner-managers, and both are consistent with the uneven digital uptake documented across smaller Australian firms (ABS 2023; ASBFEO 2023).

Several limitations qualify these conclusions. The design measured adoption intention rather than observed use, and intention is an imperfect proxy for behaviour. The cross-sectional structure precludes strong causal claims, and survey constructs were self-reported and subject to common method and social desirability biases. Finally, the sample, although diverse, was recruited partly through business networks and may over-represent more digitally engaged owners.

Conclusion

This dissertation examined the factors shaping generative AI adoption among Australian SMEs through an integrated TOE-UTAUT lens and a convergent mixed methods design. Adoption intention was driven primarily by performance expectancy, relative advantage and effort expectancy, and constrained by data governance concern and perceived cost, with the model explaining a substantial share of variance and the interviews adding explanatory depth. The Australian evidence indicates that adoption is a matter of capability and confidence rather than mere exposure: owner-managers move when the benefit is legible and the governance risk is manageable. Future research should track actual usage longitudinally and test whether targeted governance support and low-cost skills programs convert intention into sustained productivity gains, ideally drawing on national business statistics to link firm-level adoption to measured performance.

References

Australian Bureau of Statistics 2023, Business use of information technology, 2021-22, ABS, Canberra.

Australian Small Business and Family Enterprise Ombudsman 2023, Small business matters, ASBFEO, Canberra.

Braun, V & Clarke, V 2006, ‘Using thematic analysis in psychology’, Qualitative Research in Psychology, vol. 3, no. 2, pp. 77-101.

Bryman, A 2016, Social research methods, 5th edn, Oxford University Press, Oxford.

Creswell, JW & Plano Clark, VL 2018, Designing and conducting mixed methods research, 3rd edn, Sage, Thousand Oaks, CA.

CSIRO 2023, Artificial intelligence for Australia’s future, CSIRO Data61, Canberra.

Department of Industry, Science and Resources 2019, Australia’s artificial intelligence ethics framework, Australian Government, Canberra.

Fornell, C & Larcker, DF 1981, ‘Evaluating structural equation models with unobservable variables and measurement error’, Journal of Marketing Research, vol. 18, no. 1, pp. 39-50.

Hair, JF, Risher, JJ, Sarstedt, M & Ringle, CM 2019, ‘When to use and how to report the results of PLS-SEM’, European Business Review, vol. 31, no. 1, pp. 2-24.

Office of the Australian Information Commissioner 2023, Australian Privacy Principles guidelines, OAIC, Sydney.

Oliveira, T & Martins, MF 2011, ‘Literature review of information technology adoption models at firm level’, Electronic Journal of Information Systems Evaluation, vol. 14, no. 1, pp. 110-121.

Productivity Commission 2024, Making the most of the AI opportunity, research paper, Productivity Commission, Canberra.

Tornatzky, LG & Fleischer, M 1990, The processes of technological innovation, Lexington Books, Lexington, MA.

Venkatesh, V, Morris, MG, Davis, GB & Davis, FD 2003, ‘User acceptance of information technology: toward a unified view’, MIS Quarterly, vol. 27, no. 3, pp. 425-478.

Venkatesh, V, Thong, JYL & Xu, X 2012, ‘Consumer acceptance and use of information technology: extending the unified theory of acceptance and use of technology’, MIS Quarterly, vol. 36, no. 1, pp. 157-178.