I’ve sat through dozens of AI business case presentations. Most fail before they leave the room. Not because the technology isn’t promising, but because the business case reads like a vendor pitch deck rather than a serious investment proposal.

Here’s what actually works.

Start With the Problem, Not the Technology

The most common mistake I see: teams lead with the AI capability instead of the business problem. “We want to implement a large language model” tells finance nothing. “We’re losing $2.3M annually to manual invoice processing errors” gets their attention.

Your business case should answer three questions in the first paragraph:

1. What’s the specific problem costing us money?
2. How much is it costing us (with receipts)?
3. Why hasn’t this been solved before?

If you can’t quantify the problem, you don’t have a business case. You have a science project.

The ROI Model That Actually Works

Forget the hockey stick projections. CFOs have seen too many of those. Harvard Business Review recommends building a three-tier model instead:

Conservative Case (60% probability): Assume everything takes twice as long and delivers half the benefit you expect. This is your baseline.

Expected Case (30% probability): Your realistic estimate with known challenges factored in.

Optimistic Case (10% probability): The vendor’s promised outcomes. Yes, only 10%.

When I was at a Big 4 firm, we used this weighted probability model for technology investments. It’s boring, it’s conservative, and it works. Finance teams trust it because it acknowledges uncertainty rather than pretending it doesn’t exist.

Hidden Costs You’re Probably Missing

Every AI business case I review underestimates these:

Data preparation: Plan for 40-60% of your total project budget. Not a typo. Your data is messier than you think.

Change management: The technology is the easy part. Getting 500 staff to actually use it? That’s where projects die.

Ongoing maintenance: AI models degrade. Budget for retraining, monitoring, and the inevitable “why is it giving weird answers” investigations.

Integration: Your shiny new AI needs to talk to your 15-year-old ERP system. That conversation won’t be cheap.

What Finance Actually Wants to See

Based on conversations with CFOs at mid-market Australian companies, here’s their checklist:

• Payback period under 18 months for discretionary tech spend
• Clear ownership of the project and its outcomes
• Exit strategy if it doesn’t work
• Staged funding tied to milestones, not a big bang budget
• Comparison to alternatives including “do nothing”

That last point is crucial. If your business case doesn’t seriously evaluate the “do nothing” option, it looks like advocacy rather than analysis.

A Framework Worth Stealing

Structure your business case like this:

Executive Summary (1 page): Problem, solution, investment required, expected return, key risks.

Problem Analysis (2-3 pages): Deep dive on the current state, with data. Interview the people actually doing the work.

Solution Options (2-3 pages): At least three approaches, including doing nothing. Honest pros and cons for each.

Recommended Approach (3-4 pages): Your recommendation with detailed cost breakdown, timeline, and success metrics.

Risk Analysis (1-2 pages): What could go wrong and how you’ll mitigate it.

Appendices: Supporting data, vendor comparisons, technical requirements.

Red Flags That Kill Business Cases

I’ve seen all of these sink otherwise promising proposals:

• ROI calculations that assume 100% adoption from day one
• No mention of change management or training
• Vendor-supplied case studies as the primary evidence
• Vague success metrics like “improved efficiency”
• Missing operational costs after go-live
• Ignoring the opportunity cost of internal resources

The Conversation Before the Presentation

Here’s something they don’t teach in business school: the formal presentation should be a formality. Before you ever book that meeting room, you need to have individual conversations with every stakeholder.

Find out their concerns. Understand their priorities. Address their objections before they voice them publicly. By the time you present, every decision-maker should have already said “this makes sense” in private.

Yes, it’s slower. Yes, it’s political. It’s also how things actually get done in large organisations.

Final Thought

The best AI business cases I’ve seen share one quality: intellectual honesty. They acknowledge uncertainty, present realistic timelines, and don’t oversell the technology. Paradoxically, this restraint makes them more compelling, not less.

Finance teams have been burned by tech projects before. They’re looking for reasons to say no. Don’t give them any.