Generative AI and Accounting
What the best AI users do differently — and how to level up all your employees
By Nick Hallman, Zach Kowaleski, Anu Puvvada, and Jaime J. Schmidt
As organizations rapidly deploy generative AI tools, many struggle to determine whether those tools are meaningfully improving work quality, speed, and judgment. Most companies default to measuring activity — such as frequency of use or number of prompts — rather than sophistication or impact. As a result, performance gains remain uneven, and leaders lack clear guidance on what effective AI use actually looks like.
To address this gap, KPMG partnered with researchers at The University of Texas at Austin to study real-world AI use. They analyzed more than 1.4 million prompts generated by 2,500 employees over eight months, using advanced models to identify observable behaviors associated with high-impact human-AI collaboration. This allowed them to move beyond snapshots of usage and instead identify patterns that distinguished routine users from sophisticated ones.
The analysis revealed that only about 5% of employees demonstrated highly sophisticated AI use, even though nearly 90% used AI regularly. These advanced users spanned roles and seniority levels, though employees above manager level were overrepresented. The key distinction was not comfort with AI, but how it was used.
Four behaviors consistently set top users apart. First, they approached AI ambitiously, engaging in longer, more iterative interactions, writing more detailed initial prompts, and switching models or tools based on the task. Second, they treated AI as a reasoning partner rather than a one-off answer generator, actively guiding the model through role definition, examples, structured reasoning, and self-verification. Third, they delegated complex, multistep tasks by clearly defining objectives, constraints, and success criteria. Finally, they used AI as a general cognitive tool across many types of work — such as ideation, analysis, and technical problem-solving — rather than limiting it to productivity shortcuts such as drafting text.
The research also showed that frequency of use alone is a poor indicator of productivity. Junior employees were more likely to use AI for personal tasks, while senior employees applied it more strategically across a broader set of professional problems. Experience and role context shaped not just how often AI was used, but how thoughtfully it was integrated into core work.
For leaders, the implication is clear: Advancing AI maturity requires shaping habits, not just driving adoption. Organizations should define and make visible what “good” AI-enabled work looks like, invest in hands-on, scenario-based training, and set clear, role-specific expectations for AI use. Sophisticated AI collaboration is built from observable, teachable behaviors — and with the right framework, those behaviors can be scaled across the workforce.
The full version of this article was published at the Harvard Business Review site hbr.org on Mar 19, 2026. Authors Nick Hallman, Zach Kowaleski, and Jaime J. Schmidt are accounting faculty members at the McCombs School of Business. Anu Puvvada is a principal with KPMG.
About this Post
Share:
