Skip to content

AI Model Auditing

Building Trust Through Verification

Artificial Intelligence has the ability to transform how finance professionals work.

It can summarise reports, analyse data, generate financial insights, and automate repetitive tasks.

However, professional finance requires more than speed.

It requires accuracy, judgement, and accountability.

AI-generated outputs should never be accepted without review.

The ability to evaluate, challenge, and verify AI outputs is a critical skill for the AI-enabled finance professional.

Why Auditing Matters

Finance professionals make decisions involving:

  • Investments
  • Capital allocation
  • Business strategy
  • Risk management
  • Financial reporting

Errors in these areas can have significant consequences.

A financial analyst would never accept an Excel model without checking:

  • formulas,
  • assumptions,
  • calculations,
  • logic.

The same principle applies to AI.

AI outputs require professional review.

The AI Audit Mindset

The correct mindset is:

Trust the process, not the output.

AI should be treated like a highly capable junior analyst.

A junior analyst may:

  • work quickly,
  • identify useful information,
  • create a first draft,

but their work still requires review.

AI is no different.

Common AI Risks

1. Hallucinations

AI models may generate information that appears credible but is incorrect.

Examples:

  • Invented facts
  • Incorrect company information
  • Fake citations
  • Incorrect financial figures

Always verify important information against reliable sources.

2. Incorrect Calculations

AI can explain financial concepts but may occasionally make mathematical errors.

Examples:

  • Incorrect valuation calculations
  • Wrong percentage changes
  • Incorrect financial ratios

Important calculations should be independently verified.

3. Missing Context

AI may provide technically correct information but fail to understand the broader business context.

Example:

A declining margin may indicate:

  • increased costs,
  • strategic investment,
  • pricing pressure,
  • temporary disruption.

Numbers require interpretation.

4. Outdated Information

AI models may not always have access to the latest information.

Always verify:

  • financial statements,
  • market data,
  • regulatory changes,
  • company announcements.

The AI Audit Framework

A professional AI review process can follow five steps.

Step 1: Verify the Source

Ask:

  • Where did this information come from?
  • Is the source reliable?
  • Can the claim be independently confirmed?

For finance analysis, prioritise:

  • Annual reports
  • Regulatory filings
  • Investor presentations
  • Company disclosures

Step 2: Check the Logic

Ask:

  • Does the conclusion make sense?
  • Are assumptions reasonable?
  • Does the reasoning follow?

Example:

AI says:

"Revenue growth increased because profitability improved."

Question:

"Is revenue growth actually related to profitability?"

The conclusion may not follow.

Step 3: Validate the Numbers

Check:

  • Calculations
  • Percentages
  • Financial ratios
  • Valuation outputs

Example:

If AI calculates a revenue growth rate:

Revenue Growth =
(Current Year Revenue - Prior Year Revenue)
/
Prior Year Revenue

Verify the calculation independently.

Step 4: Challenge Assumptions

Every financial analysis depends on assumptions.

Examples:

  • Revenue growth
  • Margin expansion
  • Discount rate
  • Terminal growth rate

Ask:

  • Why is this assumption reasonable?
  • What evidence supports it?
  • What happens if the assumption changes?

Step 5: Apply Professional Judgement

The final responsibility belongs to the finance professional.

AI can provide:

  • Information
  • Analysis
  • Suggestions

The professional provides:

  • Context
  • Experience
  • Decision-making

AI Auditing in Financial Modelling

Financial models provide a useful example.

AI can help:

  • Generate formulas
  • Explain spreadsheet logic
  • Review model structure
  • Identify possible errors

However, a finance professional must still check:

Formula Accuracy

Does the formula calculate what it should?

Model Integrity

Do the statements balance?

Assumption Quality

Are forecasts realistic?

Business Logic

Does the model reflect how the company actually operates?

Example: AI-Assisted Company Analysis

Imagine asking AI:

"Analyse Breville Group's financial performance."

AI may provide:

  • Revenue trends
  • Margin analysis
  • Strategic observations

A finance professional should then ask:

Verification

Are these figures from the annual report?

Context

Why did performance change?

Interpretation

What does this mean for future performance?

Decision

What action should an investor or manager take?

AI Auditing Checklist

Before using AI output in professional work:

Question Check
Is the information accurate? Verify sources
Are calculations correct? Recalculate
Are assumptions reasonable? Challenge inputs
Does the logic make sense? Review reasoning
Would I present this to a senior stakeholder? Apply judgement

AI + Finance Professional Workflow

The future workflow is not:

Question → AI Answer → Decision

It is:

Question
   ↓
AI Assistance
   ↓
Professional Review
   ↓
Validation
   ↓
Decision

The value comes from combining AI speed with human judgement.

Learning Outcomes

After completing this module, you should understand:

  • Why AI outputs require verification
  • Common AI risks in finance
  • How to audit AI-generated analysis
  • How to combine AI efficiency with professional judgement

Try It Yourself — Breville Case Study

Open the Breville FY2025 Annual Report included in this repository under the Case Studies > Breville Group section.

Using your preferred AI assistant, generate an AI summary of the Breville FY2025 Annual Report.

Then audit the output by checking:

  • Are all financial figures correct?
  • Are revenue growth rates accurate?
  • Has AI invented information that is not in the report?
  • Are management comments represented fairly?
  • Are important risks omitted?

Record every error you identify.

Goal: Develop the habit of validating AI outputs before relying on them in professional finance work.

Next Steps

You are now ready to apply AI to practical finance workflows:

  • Financial statement analysis
  • Annual report analysis
  • Financial modelling
  • Valuation
  • Investment research