Most AI business cases die in the CFO’s inbox. Not because the idea is bad. Because the document reads like a vendor brochure with a price tag stapled to it.
I’ve watched this happen dozens of times. A smart operations leader gets excited about an AI use case. They put together a deck. The deck has logos, buzzwords, and a hockey-stick chart promising 40% efficiency gains. The CFO reads three slides, asks one question the deck can’t answer, and the project quietly disappears.
The problem isn’t the AI. The problem is how the case was built.
After 20 years sitting in budget meetings at IBM and Kyndryl sometimes presenting the case, sometimes watching others present it, I’ve seen the patterns. The cases that get approved share a structure. The cases that get killed share a different one.
Here’s how to write the kind that gets funded.
Start With the Cost of Doing Nothing
Every approved business case I’ve seen opens the same way. Not with the AI solution. With the problem the business is already paying for.
CFOs think in terms of money already leaving the building. If your case starts with “AI can do X,” they have to imagine the value. If your case starts with “We are currently spending $400,000 a year on Y because of Z,” they’re already nodding.
When we built the case for the customer operations transformation that ultimately delivered over $2M in documented savings, we didn’t lead with the technology. We led with the math of the current state. Incident response time. Engineer hours spent on repeat tickets. The cost of every escalation that should have been caught upstream.
By the time the AI conversation started, the audience was already convinced something had to change. The only remaining question was which option had the best return.
That’s the position you want to argue from.
Use Specific Numbers Even If You Have to Estimate Them
Vague numbers kill business cases. “Significant time savings” means nothing. “Improved customer experience” means less than nothing. A CFO has been hearing those phrases for 25 years and knows they translate to zero on the spreadsheet.
The case I’d write today for an SMB looking at an AI use case would have numbers like this:
- Current annualized cost of the problem: $312,000 (4 FTEs at 30% time on manual reconciliation, fully loaded)
- Projected reduction: 60% in year one, 75% in year two
- Net year-one savings after implementation: $147,000
- Payback period: 14 months
You don’t need perfect numbers. You need defensible numbers. If your CFO asks where the 60% comes from, you’d better have an answer even if the answer is “this is a conservative estimate based on the vendor’s documented results at three similar-sized clients, which I have verified.”
When we delivered the 38% incident detection improvement for a customer, the original business case projected 25%. We came in higher than the case. That’s the position you want to be in, under-promising and over-delivering and not the reverse.
Frame It as Risk Reduction, Not Innovation
Here’s a hard truth about CFOs: most of them don’t get rewarded for funding innovation. They get punished for funding mistakes.
The asymmetry is real. A CFO who approves ten good projects and one bad one is remembered for the bad one. So they’re not looking for the upside of your project. They’re looking for the downside.
Smart business cases acknowledge this directly. Instead of selling vision, they sell de-risking.
Frame your case in terms of:
What happens if we don’t do this in the next 18 months? Competitors are already doing it. Our manual process is fragile and dependent on three people who could leave. Our error rate is growing as volume grows.
What’s the worst-case scenario if the project underperforms? We’ve structured the engagement in three phases with go/no-go decisions. We can stop after Phase 1 with only $80K committed. The infrastructure work in Phase 1 has standalone value even if we never deploy the AI layer.
That second point is the one that closes deals. Most AI projects fail because they’re presented as all-or-nothing. The ones that get funded are the ones where the early phases create value on their own regardless of whether the AI ever ships.
Separate Infrastructure Costs From AI Costs
This is where most AI business cases fall apart under scrutiny.
The vendor quote says $200,000. The business case shows $200,000. Six months in, the project is at $450,000 and climbing because the data pipeline wasn’t ready, the integration layer needed to be rebuilt, and security review added three months no one budgeted for.
When the case includes a number the CFO knows is fake, every other number in the case becomes suspect.
The fix is to build the case in two distinct buckets. Infrastructure readiness costs: what you’ll spend whether or not you ever deploy AI. Data quality work. Integration layers. Security and governance. These have standalone value and they’re going to happen eventually. AI implementation costs: the model, the deployment, the change management, the ongoing operations.
Most SMBs I work with are surprised that the infrastructure bucket is usually 50–70% of the total. They’re equally surprised that it’s the easier number to defend, because every dollar in that bucket pays for itself even if the AI portion gets delayed or scoped down.
This is the entire reason my firm leads every engagement with an infrastructure assessment before we talk about AI strategy. Not because it’s good consulting hygiene. Because it’s the only honest way to build the business case.
Show the Operating Model, Not Just the Project
A CFO doesn’t fund projects. A CFO funds operating models.
The difference matters. A project ends. An operating model keeps running and someone has to pay for it.
Most AI business cases stop at deployment. They show implementation costs, projected savings, and a payback period. They don’t show what it costs to run the thing in year two, year three, year four.
That gap is where CFO trust dies. Because every CFO has been burned by a project that came in under budget and then quietly added $200K a year in operating cost no one mentioned.
A strong case includes the year-three operating picture. Licensing. Cloud consumption. The person or partner responsible for keeping it running. The retraining cadence for the model. The governance review cycles. The headcount implications with both reductions and any new specialist roles required.
If your numbers still pencil out after you’ve shown that picture honestly, the CFO has no real reason to say no.
The One Slide That Closes the Deal
If I had to compress everything above into a single slide, this is what it would say:
Current state cost: [specific annual number] Total investment over 24 months: [infrastructure number] + [AI number] = [total] Ongoing annual operating cost from year three forward: [specific number] Cumulative net savings over 36 months: [specific number] Payback period: [months] What we stop doing if Phase 1 underperforms: [specific exit point and sunk cost]
That’s it. No logos. No buzzwords. No hockey stick.
Every executive I’ve ever presented to has approved cases that looked like that. The ones with the gradient backgrounds and the AI-generated illustrations get polite nods and quiet deaths.
Where Most Cases Actually Get Stuck
In my work with the clients currently in our portfolio, the single most common failure point isn’t the math. It’s the answer to one question the CFO almost always asks:
“What happens to the people whose work this AI replaces or changes?”
If you don’t have a clean answer like change management plan, retraining program, role redefinition, communication strategy then your business case is dead even if every other number checks out. Because what the CFO is really asking is whether you’ve thought about what happens after the press release.
This is why the AI Change Management work is part of every Summit engagement, not an afterthought. The cases that get approved have a credible answer to that question. The cases that get killed have a slide that says “Change management to be determined.”
Don’t be that slide.
What to Do This Week
If you’re working on an AI business case right now, do three things before you put another minute into your deck.
First, write down the annualized cost of your current state in dollars. Not a range. A number. If you can’t, that’s your homework and you can’t make a financial case without a financial baseline.
Second, separate your implementation costs from your infrastructure costs. If you can’t separate them, you don’t yet understand the project well enough to ask for the money.
Third, write the one-slide summary I described above on paper. If you can’t fill in every number, that’s your gap list. Close those gaps before you present to anyone.
A CFO doesn’t want to be sold. A CFO wants to be given a defensible case they can sign their name to. Build that case, and the budget conversation gets a lot shorter.
Ready to build a business case your leadership will fund?
- Start with an AI Infrastructure Assessment: get the readiness scorecard, gap analysis, and cost separation that anchors every defensible business case. Schedule a conversation →
- Explore our services: from assessment through implementation and change management. See how we work →
- Read more on AI strategy and infrastructure: practical frameworks, no hype. Visit the blog →
Russell Love is the Founder & CEO of Summit AI Business Solutions, based in Browns Summit, NC. With 20+ years of enterprise transformation experience at IBM and Kyndryl, Russell helps businesses build the foundations that make AI actually work.