AI / Introductory

AI Agents Case Study

KPMG MADA Program Team

January 2026

Download:

Summary

This case study is designed to introduce agentic AI, as well as demonstrate how agentic AI or large language models (LLMs) can be used to complete audit-specific tasks. The materials reinforce the need for human-in-the-loop review, which includes maintaining sufficient professional judgment and reperforming procedures to evaluate the outputs of Generative AI.

Content

This multi-part case study focuses on introducing agentic AI, reviewing AI outputs, and engineering prompts to complete different audit use cases centered around risk assessment and expense testing. Students also have the option to brainstorm and create their own AI agent.

These materials are intended for use by undergraduate or graduate accounting faculty who are looking to introduce agentic AI in the context of an audit. Some prior knowledge of GenAI, as well as foundational audit knowledge, would be helpful to complete this case but is not required.

The learning objectives of this case study are as follows:

Learn about the basics of agentic AI, including the important human-in-the-loop review points when interacting with an agent and its outputs.
Improve prompting and context engineering when utilizing LLMs.
Practice reviewing both agentic AI and LLM outputs with professional judgment and skepticism, specifically to achieve appropriate and sufficient audit documentation.
Use GenAI to complete an initial business process risk assessment, as well as substantive expense testwork.
Explore the process of utilizing agentic AI for audit substantive testwork, comparing this to how the same work is completed with an LLM.
Articulate how you interact with GenAI, including explaining your logic and chain of prompts.
Use GenAI to assist with various stages of the audit, including risk assessment, controls, substantive, and audit execution.
Recognize the benefits and the risks of using this technology for staff-level audit work, as well as the relevant ethical considerations.
(If completing Part 4): Build your own AI agent to assist with a daily task.