The finance function has always adopted tools that changed the way work is performed.

For instance: Spreadsheets changed analysis, ERP systems changed transaction processing, Business intelligence tools changed reporting, & Cloud platforms changed access to information. Artificial intelligence is now changing the next layer: research, interpretation, forecasting, commentary, anomaly detection, and decision support. That is a meaningful shift. But for CFOs, the central question is not whether AI can produce a good-looking answer. It often can. The central question is whether the answer can be trusted, explained, challenged, reconciled, and used in a decision.

Finance does not operate in a world where “it sounds reasonable” is enough. A forecast affects hiring, pricing, procurement, working capital, and capital allocation. A technical accounting conclusion affects recognition, measurement, disclosure, and audit evidence. A cash-flow warning affects funding decisions. A board-pack narrative affects executive confidence.A control exception affects governance. For that reason, finance leaders should treat AI as an augmentation layer, not as a replacement for professional judgment. The practical principle is simple: Think first. Prompt second. Check before using.

This article develops a practical framework for AI-augmented finance work. The objective is not to make finance people dependent on AI. The objective is to help finance teams use AI with more speed, structure, and skepticism.

The CFO Problem-Solving: Finance is under pressure to be faster, but still accountable

Finance teams are being asked to do more than close the books. They are expected to explain performance, forecast risk, challenge assumptions, support transformation, advise the business, improve controls, and provide early warning signals. At the same time, many finance processes remain too manual:

• monthly variance commentary is often prepared late;
• forecast assumptions are not always clearly owned;
• reconciliations may depend on individual memory;
• reporting packs may explain symptoms instead of drivers;
• technical accounting research may be slow and fragmented;
• dashboards may show numbers without decision logic;
• process knowledge may sit in spreadsheets, emails, or one experienced person’s head.

AI can help with these issues. But only if the CFO is clear about the role AI should play. If AI is used casually, it creates new risks: unsupported conclusions, fabricated citations, confidentiality breaches, hidden assumptions, inconsistent logic, and overreliance by users who may not know enough to challenge the output. If AI is used properly, it can strengthen financial work by enabling better structure, faster iteration, broader challenges, and clearer documentation. That distinction matters.

A CFO framework for AI-augmented finance

The framework below adapts the “think, prompt, check” principle into a CFO operating model. It is designed for finance work such as: technical accounting research; monthly performance review; forecast and budget challenge; variance analysis; working-capital review; internal control documentation; management reporting; board-pack preparation; finance transformation; dashboard and automation design. The framework has eight steps.

Step 1: Define the decision before using AI

Human-led step

The first question should not be, “What can AI do?” The first question should be:

What decision needs to improve?

This is where many AI projects in finance start badly. Teams begin with the tool rather than the business problem. A CFO should first define the management decision or finance judgment at stake. Examples:

AI becomes useful only after the decision is clear. Otherwise, the finance team may produce faster analysis without improving management action.

Step 2: Establish the facts and classify what is unknown

Human-led first, AI-augmented second

Before prompting AI, finance must establish the facts independently. This is especially important for technical accounting, forecasting, and risk assessment. AI tools may fill gaps with assumptions if the user does not clearly identify what is known and unknown. For example, in an accounting research question, the finance team should document: relevant parties; contractual terms; transaction date; reporting date; approval date; economic substance; legal form; jurisdiction; applicable reporting framework; known amounts; uncertain amounts; missing documents; management intent; board or committee approvals; disclosure implications.

In a forecasting or performance question, the finance team should document: actual result; baseline forecast; volume movement; price movement; mix effect; FX effect; one-off items; recurring items; controllable items; business owner; source system; data cut-off.

Only after that should AI be used to challenge completeness. A useful prompt would be:

Based on the facts below, identify any missing information that would affect the finance analysis. Do not assume missing facts. Separate known facts, unknown facts, assumptions, and follow-up questions.

The critical instruction is: do not assume missing facts. That one sentence protects a lot of finance work.

Step 3: Select the correct finance lens

Human-led step

AI needs context. Without context, it may produce a generic answer. The finance leader must define the lens before asking for analysis. The relevant lens may be: IFRS accounting treatment; management reporting; tax impact; treasury and liquidity; credit risk; internal control; audit evidence; commercial margin; operational performance; board decision support; transformation business case.

For example, a revenue decline can be viewed through many lenses.

• Under management reporting, the question may be price, volume, mix, and customer behavior.
• Under IFRS, the question may involve revenue recognition under IFRS 15 if contract terms, variable consideration, rebates, performance obligations, or collectability are involved.
• Under credit risk, the same customer behavior may raise questions about expected credit loss under IFRS 9.
• Under executive decision-making, the question may be whether management should protect volume, defend price, tighten credit, or exit unprofitable customers.

The CFO must define the lens. AI can assist the analysis, but it should not decide the framework silently.

Step 4: Build a prompting plan, not a single prompt

AI-augmented step

A weak AI user asks one broad question, while a stronger finance user designs a prompting sequence. This matters because serious finance work is rarely answered properly in one prompt. It requires iteration. A practical prompting plan may look like this:

This is aligned with the CSI + FBI prompting approach in the AI-in-finance handbook:

• CSI: Context, Specificity, Instruction
• FBI: Format, Blueprint, Identity

For CFO-level work, a useful prompt structure would be:

Act as a senior IFRS finance controller supporting a CFO. 
Context:
[Insert facts, source data, jurisdiction, reporting framework, and objective.] 
Task:
Analyze the issue using the following structure:
1. Known facts
2. Unknown facts
3. Relevant finance/accounting issues
4. Applicable IFRS guidance to verify
5. Analysis
6. Alternative interpretations
7. Risks and controls
8. Recommended conclusion
9. Evidence required before final approval 
Important constraints:
- Do not invent facts.
- Identify assumptions separately.
- Do not cite standards unless they can be verified.
- Provide output in a table suitable for review.

The objective is not to make the prompt look clever. The objective is to make the output reviewable.

Step 5: Require AI to produce auditable artifacts

AI-augmented step, human-verified

Finance should not accept AI output only as prose. Prose is useful, but it is not enough. A finance-grade AI output should produce artifacts that can be inspected, tested, or reconciled. Examples:

This is one of the strongest practical principles from the AI-in-finance handbook: AI should generate artifacts that can be audited, tested, and reused. In finance, a good answer is not enough. The working must be visible.

Step 6: Verify against authoritative sources and approved data

Human-led step

This is the non-negotiable step. AI may assist research, but the finance professional must verify the output.

For accounting treatment, that means reviewing the applicable IFRS standard directly. For examples: IFRS 15 for revenue from contracts with customers; IFRS 9 for expected credit losses and financial instruments; etc.

For management reporting, verification means reconciling to approved sources: general ledger; trial balance; consolidation system; ERP report; billing system; bank statement; inventory system; operational system; approved budget; latest forecast; signed contract; board-approved plan.

For automation, verification means testing the logic: check formulas; review SQL joins; test Python outputs; reconcile record counts; compare totals to control totals; document exceptions; retain evidence.

The key question is always:

Can another competent finance professional review the work and reach the same conclusion?

If not, the work is not yet finance-grade.

Step 7: Convert the analysis into a decision-ready output

Human-led, AI-supported

Finance work is incomplete until it supports a decision. A long analysis may be necessary. But the executive output must be clear. For CFO and executive committee use, the final output should usually answer:

1. What happened?
2. Why did it happen?
3. What is the financial impact?
4. What is controllable?
5. What risk remains?
6. What decision is required?
7. Who owns the action?
8. When will we verify progress?

A decision-ready table could look like this:

This is where AI can help draft the output. But the CFO must own the final message. AI can help write. It cannot own the judgment.

Step 8: Document the AI process

Human-led step

If AI materially supports finance work, the process should be documented. This does not need to become bureaucracy. But the documentation should be sufficient for review. A simple AI workpaper can include:

This is especially important for technical accounting, external reporting, audit support, financial controls, and board-level outputs. The CFO does not need to make AI slow. But the CFO does need to make it governable.

Practical CFO checklist: when is AI output safe enough to use?

Before AI-supported finance work is used in a management decision, the CFO or finance leader should ask:

If several answers are “No,” the work is not ready. It may still be a useful draft. It is not yet a finance deliverable.

Conclusion: the CFO should use AI boldly, but not casually

AI will become part of finance work. That direction is already clear. The question is whether finance leaders will shape that adoption or allow it to happen informally through scattered tools, inconsistent prompts, and undocumented outputs.

The CFO should not resist AI. But the CFO should insist on discipline. AI can accelerate research, reporting, forecasting, reconciliation, and decision support. It can help finance teams see patterns earlier and prepare better analysis faster. But it must remain AI-augmented, not AI-dependent. The finance professional must still establish the facts, define the question, select the framework, verify the sources, challenge the conclusion, and communicate the decision. The most useful finance rule is still simple:

Think. Prompt. Check. Then decide.

That is how AI becomes a finance leadership capability rather than another uncontrolled layer of complexity.

References

User-provided source: Top 100 Tips for AI in Finance Handbook, referenced for audit-first principle, data tiering, CSI + FBI prompting, AI maturity model, governance registry, deterministic automation principle, and AI-generated auditable artifacts.

Gartner — AI in Finance: What CFOs Need to Know

World Economic Forum — AI is transforming finance, CFOs say

Deloitte — The CFO Guide to Tech Trends 2026

IFRS — IFRS 15 Revenue from Contracts with Customers

IFRS — IFRS 9 Financial Instruments

IFRS — IAS 1 Presentation of Financial Statements