AI is giving rise to a “Jagged Technological Frontier” (Dell’Acqua et al. 2023), a sentiment echoed by various speakers at the Tilburg Winter Symposium. According to this concept, deciding which tasks to use AI for is not trivial. Accountants need to think strategically for which task they use AI. AI might perform better for some tasks, such as performing a risk assessment. For other tasks with similar difficulty, like making client recommendations, AI might underperform. David Wood, a leading researcher in audit technology at Brigham Young University, estimates that AI can handle up to 80% of the work, leaving auditors to fine-tune the final 20%. This is also showcased by Maximilian Muhn, Assistant Professor at the University of Chicago, who presented his work entitled “From Transcripts to Insights: Uncovering Corporate Risks Using Generative AI,” which demonstrates the effectiveness of generative AI, especially models like GPT-3.5, in identifying and quantifying corporate risks from earnings call transcripts. The research indicates that AI-generated risk assessments provide more informative and predictive insights into firm-level volatility and investment decisions compared to traditional methods, underscoring the utility of generative AI in detecting corporate risks. Fig. 1 offers an example of a report generated by ChatGPT.

Figure 1.

A climate risk report generated via Chat GPT from an earnings call transcript.

The impact of AI goes beyond changing the task environment of accountants. The study “Machine Readership and Financial Reporting Decisions” by Cao et al. (2023) highlights the impact of AI on reducing financial misreporting. Their findings suggest that machines’ advanced capabilities in detecting complex patterns in financial data enhance reporting quality and exert a disciplinary effect on managers and their financial misreporting. This paper prompted discussions on how managers adapt their misreporting strategies in response to advanced machine detection capabilities and the subsequent impact on the overall financial reporting environment. Again, AI tools can be used to assess reporting quality, but accountants need to think about when to use them and anticipate how others will react to the use of AI.

With large language models (LLMs), the focus shifts from solely analyzing financial data to incorporating textual data, enabling a revolutionary approach of “doing mathematics with words,” as described by Rob Huisman, Quantitative Investment Expert at Robeco. This concept prompts professionals to consider specific textual data for analysis. For instance, this capability is being explored in fund management, where AI algorithms analyze CEO speech patterns to predict stock prices. To be part of the transformation, Huisman encouraged the accounting profession to consider the question: What words do you want to do mathematics with?

AI’s impact extends beyond mere number crunching. Many of our speakers gave examples of how LLMs can be co-pilots in accounting tasks, liquidity forecasts, revenue predictions, and drafting audit plans. In a broader sense, AI is bringing about what Huisman called an “AI gold rush.” Companies increasingly rely on AI for daily operational decisions, creating a dynamic environment where AI tools are shared for marketing purposes and, at the same time, guarded as proprietary intellectual property.

A rapidly improving landscape is ahead of us, signalling a future where AI is an integral and indispensable part of the accounting profession. It requires, however, careful thinking on where we can rely on machines (AI) and where professional judgement is still required.

Explainability, or the question “How does this work?” is a key factor in AI use, according to Huisman. It is vital to articulate the purpose of AI usage and have robust safety measures. Noorman stressed the importance of engaging stakeholders in the conversation, including those indirectly affected by AI. She argued that involving stakeholders early in the process and in asking questions is key to gaining acceptance for AI solutions. When doing so, the focus should be on how AI can enhance the business case rather than on technical details (Xivuri and Twinomurinzi 2023). Huisman agreed and further explained that the issue is no longer with computation (power); the important issue is the ability to explain to stakeholders what the algorithm is performing and what parameters influence the outcome. Put it differently: to fully benefit from the improvements in task accuracy and efficiency, there is still a need to upskill to explain and articulate the work of AI. While upskilling is necessary to stay ahead of the competition, skills such as asking the right questions, negotiating outcomes, and being sceptical are and will remain relevant.

Möller emphasized the pivotal role of human accountants in incorporating AI’s power. He noted that while AI can automate many tasks, critical insights and final decisions should ideally be overseen by skilled professionals, placing the human accountant at the center of AI integration. The synergy of human-machine interactions lays the groundwork for tasks like automated data preparation and initial forecasts. However, the critical phase is the accountant’s intervention for validation and adjustment. This involves scrutinizing the reasonableness of forecasts and, if necessary, altering parameters—decisions that rest with human professionals, as Möller explained. He also suggested that implementing frameworks governing AI use in accounting is highly dependent on individual firm characteristics, with management control emerging as a key area for such implementations. Moreover, organizational structures often play a critical role in how companies adapt and integrate AI. While companies may not struggle with data, adapting leadership and accountability across different business units is essential. He envisions that the challenge for many established companies is not just in adopting new technologies, like LLMs, but in effectively integrating them into their existing structures, often leading to complex management control challenges.

In that regard, Maurice Op Het Veld, Global Lead Technology & Digital Audit Partner at KPMG, shared five success criteria for AI implementation that were identified from surveys of AI implementation leaders and laggards:

• establishing a data strategy within the organization;
• measuring AI’s performance and value;
• securing full executive commitment;
• combining internal data with external sources for broader insights; and
• demonstrating algorithm fairness to build trust in algorithms.

These criteria provide a framework for successfully implementing and leveraging AI in accounting, ensuring its beneficial integration into the field.