Questions boards should ask before approving an AI budget

Most AI budget proposals are written by the people who want the budget approved. That is not a criticism — it is simply a reason boards need their own short list of questions, independent of whoever is presenting. The goal is not to slow good ideas down. It is to make sure the organization is buying a business outcome, not a piece of technology.

1. What management decision does this actually improve?

Every proposal should be traceable to a specific decision that gets made faster, more accurately, or more consistently as a result — a pricing decision, a collections decision, a staffing decision. If the answer is a general claim about "efficiency" or "innovation" with no named decision attached, the use case has not been prioritized properly yet.

2. What did the diagnostic actually find?

A credible proposal is grounded in a review of current processes, systems and data quality — not a vendor demonstration. Ask to see the underlying assessment: which processes were reviewed, what gaps were found, and why this use case ranked above the alternatives.

3. Who owns the risk, and where is the human still in the loop?

For any use case touching customer decisions, financial reporting, credit, or personal data, the board should be able to name the specific point where a human reviews or approves the output before it takes effect. "The system is accurate" is not a control. A named reviewer, a defined escalation path, and an audit trail are.

If nobody in the room can describe the human-in-the-loop step in one sentence, the control has not been designed yet — only the automation has.

4. How will we know if it worked?

Adoption, performance improvement and return on investment should be measurable from the outset, with a review date on the calendar — not "we'll know it when we see it." Ask what will be reported back to the board, and how often.

5. What happens to the people currently doing this work?

Automation changes roles before it changes headcount, usually. Boards should ask how affected teams will be trained and redeployed, and whether that training is funded as part of the same proposal — not treated as a separate, optional line item.

6. What is the smallest version of this we could test first?

A well-prioritized use case can usually be piloted in a contained way before a full rollout. If the proposal only exists as an all-or-nothing programme, that is worth questioning on its own.

The pattern underneath these questions

Each of these questions is really the same question asked six ways: does this proposal give management Visibility into what is happening, Control over how it happens, and Intelligence to decide what to do next — or does it just introduce a new piece of software? Proposals that can answer that clearly, specifically, and with a named owner tend to be the ones worth approving.