Twelve Questions CFOs Should Ask About AI Spending

Last month, I watched a chief technology officer (CTO) request another $4 million for artificial intelligence (AI) tools.

The chief financial officer (CFO) asked one question: “What did the first $2 million produce?”

The room went quiet.

Both executives needed the same thing: a framework that connects AI spending to financial results, operating changes, and risk controls.

The following 12 questions give a CFO a practical starting point for that review.

The 12 questions

1. What is the investment thesis and the three-year P&L forecast? Classify the goal as cost reduction, new capability, or new revenue. Define the expected timing, cost, return, and stop condition.
2. Where will productivity gains appear? Choose a measurable result, such as faster delivery, more product capacity, or lower operating cost. Calculate the net present value for each option.
3. Do compensation plans reflect changes in work and skills? Measure how AI tools affect output, role scope, and retention. Adjust compensation only when the evidence supports a change.
4. Can the data answer important business questions with citations? Test revenue, churn, and gross-margin questions. Do not fund dependent AI work until the data meets defined accuracy and response-time targets.
5. Which AI asset retains value when models change? A governed, queryable data layer can preserve corrections, labels, and business context. Keep this asset under company control.
6. Which decisions can agents make without human review? Document the authorized actions, limits, and escalation paths. Include pricing changes, refunds, support escalations, and code changes where applicable.
7. Where are the evaluation controls? Require an evaluation suite before production. Apply control rigor that matches the financial and customer risk. Include the chief audit executive in the review.
8. Has the company ranked the AI portfolio by return on investment? Use one scoring method for labor savings, revenue, risk reduction, and cost. Fund or stop projects from that evidence.
9. Is engineering cycle time improving? Measure the time from an approved idea to a merged pull request. Compare the result with a baseline and an agreed target.
10. Has the company priced the AI security risk? Review prompt injection, unauthorized AI use, agent data exfiltration, and model supply-chain risk. Check the exclusions in cyber-insurance policies.
11. Can the finance team calculate AI gross margin? Attribute model, infrastructure, vendor, and review costs to each feature. Track vendor concentration and changes in unit cost.
12. Are leaders accountable for adoption and results? Measure use in approved workflows, business outcomes, control failures, and staff retention. Connect compensation only to results leaders can control.

What changes when this works

A CFO does not need to become a model expert. The CFO must ask clear questions until each answer connects to money, operations, or risk.

Effective reviews have two common features. Each AI program has its own P&L. Each program also has an accountable business owner.

This structure changes the approval discussion. The executive team reviews an operating case, not a general AI budget.

If the team cannot explain the case in a board review, the team must revise or stop the program.

Where to start

Start with the question that has the least complete answer. Assign an owner and a review date.

A CTO can use the same list before the finance review. A prepared answer makes the investment easier to assess and defend.

Which question would you add to this review?