Core Principle
Do not delegate a higher level of authority until the evidence and controls support reliable performance at that level.
Purpose
A structured review worksheet for determining the maximum level of AI authority a business process is ready to support, based on evidence, testing, permissions, segregation of duties, monitoring, escalation, and accountability.
Who This Resource Is For
This practice guide is designed for accounting, finance, risk, internal-control, and business teams evaluating whether a proposed AI workflow is ready to retrieve, draft, recommend, decide, or execute.
When to Use It
Use the guide before delegating authority to an AI system, agent, application, or vendor-supported workflow, especially when the process affects financial reporting, customers, regulatory obligations, sensitive data, approvals, transactions, or operational decisions.
How to Use This Guide
- Document the business use case and consequence of failure.
- Identify the highest Reliability Ladder level the proposed workflow requires.
- Review evidence from lower levels before approving higher authority.
- Record limitations, required controls, unresolved issues, and human review points.
- Approve a maximum delegation level rather than treating AI use as simply approved or rejected.
- Reassess when the model, data, prompts, tools, integrations, business rules, or organizational risks change.
What Is Included
Use Case Profile
Define the business activity, owner, system, primary user, current process, proposed AI role, and consequence of failure.
Initial Risk Classification
Classify the use case as low, moderate, or high based on reversibility and consequence.
Reliability Ladder Review
Work through Retrieve, Draft, Recommend, Decide, and Execute before approving a maximum delegation level.
Data, Test Design, and Evidence Review
Evaluate representative testing, prompt sensitivity, repeatability, version documentation, validation, and severity of failure.
Delegation Decision
Record the approved authority ceiling or explain why the proposed level is not approved.
One-Page Readiness Check
Use a concise readiness checklist covering purpose, evidence, reliability testing, authority, controls, and accountability.
Connected Learning Experience
Read the article to understand the idea. Explore the framework to organize your thinking. Apply the Practice Guide in real finance and accounting work.
Learning Note and Limitations
This guide is an original educational resource developed from Valentina DuPont’s professional perspective as a CPA and her personal interpretation of professional learning. It is not an accounting, auditing, tax, legal, investment, compliance, cybersecurity, model-risk, software-engineering, or professional standard. It does not establish whether a specific AI use is legally permissible or technically secure. The appropriate review must be tailored to the use case, data, authority, risk, technology, organization, and applicable requirements.