There is a four-question framework that separates AI investments that deliver measurable return from those that quietly become expensive proof-of-concepts. If you are presenting an AI investment to your CFO or board in 2026, this framework is the difference between approval and another quarter in slide-deck purgatory. Here it is.
Why do 95% of enterprise AI projects fail to prove ROI?
According to MIT and Futurum Group research cited across the industry in 2026, roughly 95% of generative AI projects fail to deliver measurable financial impact. The cause is rarely the technology. It is that organisations launch AI without establishing a financial baseline, then cannot prove what changed after deployment. Without a baseline, ROI becomes an opinion, not a number.
This matters because the AI conversation has shifted upmarket. Five years ago an AI pilot could be funded by a department head's discretionary budget. In 2026 the spend has moved into capital review, and that means your CFO is asking three questions a marketing-flavoured pitch cannot answer.
What is the total cost over three years? Not the licence fee. The full cost stack.
What measurable business metric will move? Not productivity in the abstract. A line in the P&L.
What is the baseline today? Not last quarter's narrative. A documented number.
What belongs in an AI business case your CFO will approve?
A board-ready AI business case has four required components: a three-year total cost of ownership figure, a defined target metric mapped to the financial statements, a documented pre-deployment baseline, and a quarterly measurement plan. Anything less is a proposal. Anything more becomes noise the finance team will ignore.
The Futurum Group's 1H 2026 enterprise buyer survey found that direct financial impact, defined as revenue growth plus profitability improvement, jumped to 21.7% as the primary ROI metric finance teams will accept, while productivity gains fell to 18%. Translation: the era of "AI saves us hours" being enough is over. Your CFO wants the hours converted into either revenue, cost reduction, or risk avoidance, with the conversion logic shown.
The strongest business cases name the exact P&L line that will move and by how much. For a 200-person professional services firm in Hong Kong, that might read: "Reduce annual external translation spend from HK$1.8M to HK$600K by deploying AI translation for internal-grade output, with human review only on client-facing documents."
How do you calculate the true total cost of an AI investment?
Most AI business cases underestimate total cost by 40-60% because they only count licence fees. The full cost stack across three years includes implementation, licensing, internal engineering, change management, ongoing model evaluation, and the inference costs that scale with usage. CFOs see this gap immediately, and it kills credibility on the spot.
According to Microsoft's enterprise AI research published in 2026, the licence cost of a tool like Microsoft Copilot is typically 25-35% of the three-year total cost. The remaining 65-75% sits in categories most pitch decks omit.
The categories that consistently get missed include: integration engineering hours, prompt and workflow design, internal training rollouts, model evaluation tooling, governance review cycles, retraining when source data shifts, and the inference cost premium for high-volume use cases. A Hong Kong financial services firm running 50 concurrent AI-assisted workflows can see inference costs alone reach HK$300K-HK$500K annually before licensing.
The fix is a three-year TCO table with every category named and either costed or labelled "not applicable". Empty rows force a conversation. Hidden rows kill the case.
What are the four layers of AI ROI measurement?
The four-layer ROI model used by leading enterprise AI programmes maps activity through to financial outcome. Layer one is adoption: are people using the tool? Layer two is task-level impact: does it complete work faster or better? Layer three is process-level impact: does it move a business metric? Layer four is P&L impact: did revenue, cost, or risk actually change?
The CFO conversation that keeps AI budgets funded requires evidence from at least layer two, and ideally layer three. Layer one metrics, such as monthly active users or licences consumed, are necessary but not sufficient. According to McKinsey's 2026 State of AI survey, enterprises that report meaningful ROI are 4.2 times more likely to measure at layer three than peers who stalled at layer one.
What this means in practice: if your AI pilot dashboard only shows adoption metrics six months in, your CFO is correct to be unconvinced. Build the upper layers into the plan from day one, not as a retrospective exercise.
How do you establish a pre-deployment baseline?
A pre-deployment baseline is a documented snapshot of the metric you intend to move, captured before the AI tool is introduced. Without it, every post-deployment number is comparing against a story rather than a fact. The baseline takes two to six weeks to capture properly and is the single highest-leverage step in the entire framework.
For a target metric like "reduce average customer service response time from current state to under four hours," the baseline requires the actual current state: time-stamped response data from the previous 90 days, segmented by query type, channel, and customer tier. Anecdotes are not baselines. Last quarter's narrative is not a baseline.
Hong Kong enterprises often skip this step because the data is fragmented across email systems, ticketing platforms, and CRM records. The skipping is exactly why so many AI cases collapse under CFO questioning six months later. Better to spend the four extra weeks now than to defend a pilot with no proof.
What mistakes sink most AI business cases in finance review?
The three mistakes that kill AI business cases in finance review are inflated productivity claims with no time-to-value evidence, missing cost categories that surface post-deployment, and ROI projections that confuse adoption with impact. Each one signals to a CFO that the team has not done the underlying work, and finance review is where that gets exposed.
Inflated productivity claims typically come from vendor case studies that assume best-in-class implementation. A vendor saying "customers see 35% productivity gain" is meaningful only if your organisation looks like those customers. Apply a 50% discount factor to vendor-supplied numbers unless you have a like-for-like comparable.
Missing cost categories show up in month four when the integration partner submits a change order, when the governance team requests dedicated evaluation infrastructure, or when inference bills run higher than the licence. Each of these is foreseeable. Each one shows up in CFO post-mortems.
Confusing adoption with impact is the most common failure mode. A 90% licence utilisation rate proves people are logging in. It does not prove anything moved in the P&L. The CFO knows the difference.
How should you present your AI business case to the board?
A board-grade AI business case fits on one page and answers three questions before the slide turns: what is the financial commitment, what is the expected return, and what is the risk if it underperforms. The supporting detail belongs in the appendix. The headline page is what gets discussed.
The single most effective format is a four-quadrant view: top-left, the investment ask; top-right, the projected three-year P&L impact; bottom-left, the named target metric and current baseline; bottom-right, the downside scenario and exit plan. Boards do not approve AI cases. Boards approve commercial cases that happen to use AI.
According to Harvard Business Review's 2026 analysis of enterprise AI governance, the cases that get funded share one trait: a named executive accountable for the outcome, not the deployment. Deployment is a project management metric. Outcome is what the board funds.
The strategic takeaway: building an AI case that survives finance review
The framework that separates approved AI investments from quietly shelved ones in 2026 is not technical sophistication. It is financial rigour applied before the pilot begins. A three-year TCO with every cost category named, a target metric mapped to the P&L, a documented baseline, and a quarterly measurement plan. That is the entire framework, and it is what your CFO is checking for the moment your slide opens.
Hong Kong enterprises have a real advantage here. The market is small enough that benchmarks travel quickly between peer organisations. The leaders building credible AI investment cases in 2026 are the ones doing the unglamorous baseline work in the first six weeks, not the ones rushing to deploy.
We understand the cold edges of AI and the hard parts of your work, and UD has walked with Hong Kong enterprises for twenty-eight years, making technology a partnership with warmth.
Ready to build an AI investment case your CFO will approve?
Now that you have the framework, the next step is establishing the baseline that will carry your case through finance review. Our team will walk you through every step, from AI readiness assessment to baseline measurement, ROI modelling, and board-grade business case construction. Twenty-eight years of partnering with Hong Kong enterprises, with you the entire way.
