Corporate enterprise AI deployments are largely functioning as glorified search engines rather than workflow automation engines. An empirical study by University of Texas at Austin researchers Nicholas J. Hallman, Zachary T. Kowaleski, Jaime J. Schmidt, and KPMG's Anu Puvvada analyzed 713,564 prompts from nearly 4,000 back-office workers across KPMG LLP over eight months in 2025.
The findings deflate enterprise software marketing: prompting sophistication showed zero organic growth over time, and formal corporate training initiatives yielded no lasting improvements in prompt engineering or structured workflow automation. While enterprises funneled an estimated $37 billion into generative AI deployments in 2025, back-office users plateaued at shallow, single-turn interactions. Complex chaining and advanced prompting remained confined to senior domain specialists inside Strategy, Digital Innovation, and Project Management.
Blanket enterprise seat licensing without deep operational restructuring offers negative ROI. Leadership must audit departmental token consumption and prompt complexity profiles before signing renewals, treating generative models as specialized infrastructure rather than self-adopting office tooling.