Challenges and potentials of AI in the tension between automation and the advisory role – An empirical study among austrian tax advisors and auditors

Abstract

This diploma thesis examines the challenges and opportunities presented by artificial intelligence in the fields of tax consulting and auditing in Austria. It focuses in particular on the balance between increasing automation on the one hand and the continued need for human assistance on the other. The thesis begins with three central questions designed to guide the research process: To what extent is AI currently used for routine tasks and what do practitioners have to say about this? What is likely to be technically replaceable in the future, and what is not? Furthermore: To what extent does the use of such systems influence the relationship with clients and their trust? Methodologically, the thesis proposes measuring AI use not as a binary variable but along a distinction between AI-assisted support and AI-driven process automation, and shows that binary measurement systematically overstates automation depth. As a second contribution, it identifies an epistemic dimension of risk perception, namely the concern that AI output may be incorrect without this being detectable without domain expertise, which established acceptance models do not represent.Methodologically, an exploratory-sequential mixed-methods design (QUAL $\rightarrow$ quan) is employed. First, semi-structured interviews are conducted with five experts. The information gathered from these interviews feeds directly into the subsequent online survey of 37 individuals from the field of public accounting. Finally, the two data streams are linked using systematic triangulation.The results are differentiated. Although $54\,\%$ of respondents rate document capture and automated posting as already highly automated, the handling of verification tasks such as journal entry testing remains largely manual only $14\,\%$ of respondents use advanced automation here. This indicates significant room for improvement. Discussions and interviews revealed a key insight: AI-supported assistance is fundamentally different from fully automating an entire process via artificial intelligence. It is precisely this distinction that explains why, despite reported technology use in $87\,\%$ of cases, true automation is rarely achieved in practice.When assessing risks, concern about hallucinations is the most frequently cited, it stands at $78\,\%$, surpassing IT security ($73\,\%$). These two are followed by data protection, cited by just under six in ten respondents, and liability issues, mentioned by nearly half of those surveyed. At the same time, expectations are high: around three-quarters of all participants believe that the proportion of IT-related activities will increase in the future. In their own assessment, knowledge of the fundamentals of data analysis scores the lowest here, the mean is $\overline{x} = 2.95$ points. The majority of participants advocate for the establishment of clear professional standards for working with AI\@.This study makes three key contributions: The first contribution lies in clearly distinguishing between AI-supported assistance and fully automated processes. Using empirical data, it demonstrates for the first time in the Austrian context the extent to which hallucinations generated by AI systems are perceived as a risk. Furthermore, it identifies a lack of foundational knowledge regarding data analysis. Building on this, the study develops concrete recommendations for action, for tax advisory and audit firms, educational institutions, and the Chamber of Tax Advisors and Auditors (KSW).