There is a four-part framework that separates AI proposals a CFO approves from the ones that quietly die in a follow-up meeting. It is not about better slides or a bigger vision. It is about giving the person who controls the budget a number they can defend to the board. Here it is.
The stakes are higher than most proposals acknowledge. According to Gartner, worldwide AI spending is on track to reach roughly US$2.5 trillion in 2026, yet only about a quarter of AI initiatives deliver the ROI expected of them. The gap is not a technology problem. It is a business-case problem.
Why do most enterprise AI business cases fail to get approved?
Most AI business cases fail because they promise transformation without a defensible number. A CFO cannot approve "efficiency" or "innovation". They can approve a specific cost reduced, a specific hour saved, or a specific revenue enabled, with a timeline. Vague upside reads as risk, and risk without a number does not get funded.
The data explains the caution. Gartner reports that roughly 20% of AI initiatives fail outright, and only about 28% fully meet ROI expectations. A CFO who has read those numbers is not being difficult. They are pattern-matching your proposal against a landscape littered with expensive pilots that produced a slide deck and nothing else.
What is the four-part framework for an AI business case?
The framework has four parts: scope the value, model realistic payback, name the metrics you will report, and de-risk the pilot. Each part answers a question a CFO will ask before signing. Skip one, and the proposal has a hole a finance team will find in the first meeting.
The logic is sequential. You cannot model payback until you have scoped value, cannot pick metrics until you know the value, and cannot de-risk what you have not defined. Worked in order, the four parts turn an aspiration into an investment memo, which is the only document a CFO actually approves.
How do you scope the value of an AI investment?
You scope AI value by tying it to one specific, measurable business process, not a department-wide ambition. Pick a task with a known cost: hours spent, error rate, response time, or headcount pressure. A narrow, quantified target is far more approvable than a broad promise to "transform operations".
Consider a concrete example. Rather than "AI will improve customer service", the scoped version reads: "Our team spends 400 staff-hours a month answering repetitive policy questions; an AI assistant grounded in our own documents can absorb an estimated 60% of that volume." Now the CFO sees a number, a source, and a boundary.
McKinsey's research reinforces why scope matters. It found that only around 21% of gen-AI adopters fundamentally redesigned the underlying workflow, yet that group was roughly 3.6 times more likely to report meaningful EBIT impact. Value comes from redesigning a process, not from bolting AI onto a broken one.
What is a realistic AI payback period for enterprises?
A realistic enterprise AI payback period is two to four years for a substantial deployment, not the seven to twelve months leaders often assume. Deloitte's survey of over 1,800 executives found AI payback typically runs three to four times longer than conventional technology, and only about 6% see returns inside one year.
This is the single most important honesty in the whole business case. A proposal that promises payback in six months will be believed by no experienced CFO and will damage your credibility when it slips. A proposal that models a realistic two-to-three-year payback, with a smaller quick win in the first year, reads as competent rather than naive.
The strategic move is to stage it. Fund a narrow pilot with a fast, modest return to prove the mechanism, then use that evidence to justify the multi-year investment. CFOs fund proven mechanisms far more readily than promised ones.
Which metrics should you report to prove AI ROI?
Report metrics that map directly to money or risk: cost per transaction, hours reclaimed, error rate, cycle time, and adoption rate. Vanity metrics like "queries answered" prove activity, not value. The right metric is one the CFO can trace to the P&L or to a reduced liability.
Adoption rate deserves special attention because it is the metric most projects ignore and most failures share. An AI tool used by 12% of the team delivers 12% of its modelled value, regardless of how good the technology is. Reporting adoption alongside outcomes tells the CFO whether the investment is actually landing.
The reason this matters now: Gartner notes that only about 29% of executives can confidently measure their AI ROI at all. Simply arriving with a defined measurement plan puts your proposal ahead of the majority that cannot answer the question.
How do you de-risk an AI pilot before asking for full budget?
You de-risk an AI pilot by making it small, time-boxed, and reversible, with a defined success threshold agreed before it starts. A pilot that cannot fail cheaply is not a pilot; it is a commitment in disguise, and CFOs recognise the difference immediately.
Three practical guardrails de-risk most pilots. Set a fixed budget and end date so the spend cannot drift. Define the success metric and threshold in advance, so the decision to scale is evidence-based rather than political. And confirm the data and integration reality up front, because legacy-system friction is where budgets quietly overrun.
This is also where a CFO's real question surfaces: not "will it work" but "what happens if it doesn't". A business case that answers that question honestly, with a capped downside, is far easier to approve than one that pretends failure is impossible.
What does an approvable AI business case look like in Hong Kong?
An approvable Hong Kong AI business case pairs a scoped local use case with realistic payback and a named data-governance answer. Because sectors like financial services operate under HKMA expectations and the PDPO, a CFO will ask about data handling before ROI. Answer it in the memo, not under questioning.
A strong local example: a mid-market professional services firm scopes AI to document drafting, models a two-year payback staged behind a three-month pilot, commits to adoption and cycle-time metrics, and addresses client-data confidentiality up front. That memo gets funded because every question a Hong Kong CFO would raise is already answered on the page.
The contrast is instructive. The same firm asking for "budget to explore AI" gets a polite deferral, because exploration is a cost centre and a scoped, de-risked, measurable investment is a business decision.
The strategic takeaway
The best AI business case is not the most ambitious one. It is the most defensible one: a scoped use case, a realistic multi-year payback with a fast first win, metrics that map to money, and a pilot that fails cheaply if it fails at all. That is the document that separates leaders who get budget from those who get deferred.
Building it well is easier with a partner who has sat on both sides of that table. We understand AI. We understand you. With UD by your side, AI never feels cold, and after twenty-eight years helping Hong Kong enterprises turn technology into results, we know a funded AI strategy is built on honest numbers, not big promises.
Ready to build a business case your CFO will approve?
Now that you have the framework, the next step is finding your highest-value, lowest-risk entry point and putting real numbers behind it. We'll walk you through every step, from AI readiness assessment to scoping the pilot, modelling payback, and defining the metrics your board will trust, with twenty-eight years of enterprise experience behind you.
