Using AI Effectively in Financial Modeling: Modeler Responsibility from Problem Definition to Decision Support (Journal of Accounting and Finance)

Summary

• The article presents the FRAME, DESIGN, BUILD, VERIFY framework for using AI in financial modeling without surrendering ownership of the model.p.1
• VERIFY applies three levels of review: technical, financial and decision verification.p.1
• The framework adds three requirements for AI-assisted work, including defined AI roles and evidence of modeler ownership.p.2
• Data sourcing runs across all stages rather than being a separate fifth step.p.3
• Four roles are distinguished: the modeler, the reviewer, the decision-maker and the AI tool.p.8
• The framework is illustrated through private wealth management, capital budgeting and financial forecasting.p.1

Key insights

1. The larger risk of AI in modeling is the illusion of correctness, where a technically sophisticated workbook answers the wrong question.p.2
2. AI-generated formulas should be treated as drafts rather than completed work.p.3
3. AI should flag a gap rather than fill it with an unsupported estimate unless the modeler has approved an estimation method.p.5
4. A model can be technically correct and still be financially wrong or irrelevant to effective decisions.p.7
5. If verification shows the architecture no longer represents the decision, the modeler should return to DESIGN rather than repair isolated formulas.p.8
6. If prior AI assistance was not documented, material sections of an inherited model should be treated as unverified.p.8
7. Documentation is evidence of process, not proof of understanding.p.9