AI in highly skilled professions: Employers plan adaptation rather than replacement

A new study by researchers of the TRR 266 Accounting for Transparency sheds light on how employers perceive the growing influence of artificial intelligence on skilled professions. The study, entitled “Beliefs About Bots: How Employers Plan for AI in White-Collar Work” provides the first randomized-experimental, evidence-based insights on how corporate management adapts its expectations and strategies when confronted with credible information about automation risks.

The study focuses on the service sector, particularly tax consulting firms in Germany. It examines how information about AI-supported automation influences companies’ beliefs, hiring decisions, and future plans.

Employers underestimate automation risks in highly skilled professions

The study challenges a widespread assumption: that office as well as knowledge work are largely immune to automation. In a randomized information experiment, the researchers show that employers systematically underestimate the extent to which AI technologies can already automate professional, cognitive, and analytical tasks today.

Figure: Distribution of subjective assessments of the automation potential of various tax-related activities compared to estimates by the IAB Job Futuromat. Source: Brüll, Mäurer, and Rostam-Afschar (2025).

Before the information intervention, most companies assumed that automation risks primarily affected simple administrative tasks, payroll accounting, or tax filing. However, after being confronted with expert estimates of the capabilities of modern AI systems, they significantly adjusted their expectations upward. This effect was particularly strong for activities characterized by repetitive analytical tasks, such as data entry, accounting and standardized reporting.

The impact of AI on highly skilled professions is therefore greater than many employers assume. Even activities that require judgment, interpretation, or specific expertise can be partially automated. Nevertheless, these changed risk assessments did not lead to immediate changes in hiring behavior. Many companies view the introduction of AI as a gradual adaptation process rather than a sudden upheaval.

Information changes expectations, not short-term personnel planning

The central research question of the experiment was whether updated beliefs about automation risks influence short-term employment strategies. To this end, randomly selected companies were provided with reliable data on the proportion of activities in their industry that are potentially automatable.

After the intervention, the perceived probability of automation increased, while hiring plans remained unchanged in the short term. This shows that employers are increasingly aware of the potential of AI, but do not immediately respond with staff cuts or hiring freezes. Instead, they focused their attention more on restructuring job content and training their employees.

It is noteworthy that the updated beliefs about automation were accompanied by higher expectations for productivity and profitability—but not by rising wage expectations. This suggests that employees may only benefit partially from the expected efficiency gains. The researchers see this as a possible early sign of growing inequality within companies, for example, between employees whose activities are supplemented by AI and those whose tasks could be replaced.

In addition, the new information triggered forward-looking adaptation strategies. Many companies planned to introduce training in data analysis, AI system management, and digital compliance tools. Others intended to retrain routine workers for human-AI collaboration tasks. These findings point to a shift in attitudes: AI is increasingly perceived as a strategic resource rather than a threat.

Automation expectations boost productivity optimism but harbor inequality risks

The study shows that changing beliefs about automation go hand in hand with growing optimism about business development. After the information intervention, employers expected higher efficiency and better financial results from the use of AI.

However, this optimism is accompanied by selective adaptation. Companies expect higher returns with leaner or restructured workforces, while wage expectations remain largely unchanged. This decoupling of productivity and wage growth could further exacerbate existing income disparities in knowledge-intensive service sectors.

The introduction of AI in the office sector is more likely to reconfigure job content than to displace entire professions. Tasks that require human judgment, contextual understanding, and trust will remain central, while repetitive analytical tasks will increasingly be automated. This hybridization of job roles is already happening in areas such as tax consulting, legal services, and financial services, where AI systems perform data-intensive tasks and humans interpret and review the results.

Figure: Estimated automation potential of various activities in tax consulting.
Source: Brüll, Mäurer, and Rostam-Afschar (2025).

The study shows that employers’ responses to automation risks depend not only on technological possibilities, but also on the accuracy and credibility of the available information. Companies that receive structured, evidence-based assessments rather than alarmist forecasts adjust their expectations rationally and prepare objectively for the integration of AI.

Long-term implications for labor markets and policy

The findings have far-reaching implications for labor markets and economic policy. If employers continue to underestimate the transformative impact of AI, the adjustment process in many industries could be delayed, resulting in skills gaps and adaptation problems.

The researchers argue that policy measures should aim to align employers’ perceptions with technological progress. Targeted information and training initiatives can enable companies to invest in retraining and workplace design at an early stage. In addition, the researchers emphasize that continuous learning systems are crucial for preparing employees for an AI-supported working environment. Smaller law firms in particular use generative AI significantly less often than larger ones, which is likely due to limited resources, lower economies of scale, and a lack of technical expertise.

Figure: Use of generative AI in companies by number of employees.
Source: Brüll, Mäurer, and Rostam-Afschar (2025).

The study provides new empirical evidence that not only innovation but also information determines how societies respond to AI. By identifying the formation of beliefs as a central adaptation mechanism, the study describes automation as a social, cognitive, and strategic process – not just as technological change.

The authors conclude that the rise of AI in skilled professions will not be characterized by sudden job losses, but rather by gradual task shifts, role developments, and organizational adjustments. In the long term, differences in the perception and implementation of automation information could lead to some industries adapting smoothly, while others suffer more from disruption.

To cite this blog

Brüll, Eduard, Mäurer, Samuel, and Rostam-Afschar, Davud (2025). AI in highly skilled professions: Employers plan adaptation rather than replacement. TRR 266 Accounting for Transparency Blog. https://www.accounting-for-transparency.de/ai-in-highly-skilled-professions/