What the Best AI Users Do Differently — and How to Level Up All of Your Employees

About the study

The research team — Nicholas Hallman, Zachary Kowaleski, and Jaime Schmidt of The University of Texas at Austin, together with Anu Puvvada of KPMG LLP — analyzed 713,564 employee prompts and the corresponding large language model responses from nearly 4,000 KPMG back-office employees across 15 functional areas over eight months in 2025. Because the employees studied work in IT, marketing, accounting, finance, and other administrative functions that exist in most large organizations, the findings speak to managers well beyond a single firm.

The team measured how sophisticated each employee's generative AI use was along three dimensions: whether the employee applied the tool across a broad mix of tasks, employed deliberate prompting techniques, and provided clear and specific instructions.

What the best AI users do differently

1. Sophistication rises with seniority. Senior employees used generative AI more skillfully than staff, consistent with domain expertise complementing AI capabilities — experienced professionals are more practiced at specifying objectives, reviewing intermediate work, and pushing for the output they want.
2. Function matters. Sophistication was highest in Strategy, Digital Innovation, and Project Management — groups focused on firmwide initiatives and organizational change — while Accounting & Finance lagged on every measure, suggesting that how well AI is used depends partly on whether the work itself calls for open-ended, iterative problem solving.
3. The best users are ambitious and persistent. Longer initial prompts (asking the tool to do more) and more iteration within a conversation (refining and following up) consistently accompanied more sophisticated use — easy-to-observe signals managers can track without reading transcripts.
4. Sophistication is hard to change. Use did not become more sophisticated over the eight-month window, and formal AI training was associated with a bump only in the month the training was completed — not in the months that followed. One-time training, on its own, is unlikely to produce lasting change.

How to level up all of your employees

The study gives managers a customizable process for measuring how well — not just how often — employees use generative AI, along with benchmarks from a firm that has already achieved widespread adoption. The results caution against relying on one-off training sessions and point instead toward ongoing reinforcement, encouraging ambitious and iterative use, and recognizing that the same tool fits some functions' work far more naturally than others'.