Most AI projects stall before they start, not because the technology doesn't work, but because no one has made a credible argument for the money. Here's how to build one that actually gets a yes.
Why most AI proposals get rejected
The typical AI pitch walks into a leadership meeting with a lot of enthusiasm and not much math. “Competitors are doing it,” “it'll make us more efficient,” “we need to stay current”, these are feelings, not business cases. A CFO's job is to say no to vague spending, and a vague pitch makes their job easy.
The fix isn't a bigger slide deck. It's a tighter argument, focused on three things your finance team actually cares about: what the problem costs you now, what the proposed solution will demonstrably return, and what happens if you do nothing.
Start with what it costs you today
Before you can argue for AI investment, you need to quantify the problem you're solving. This sounds obvious, but most proposals skip it. They jump straight to “AI will save time” without ever establishing how much time, what it's worth, and whether the math adds up.
Identify a specific, measurable workflow. Not “customer service”, something more specific, like “manually routing and categorizing inbound support tickets.” Then estimate:
- How many hours per week does this consume across your team?
- What does an hour of that labor cost (fully loaded, including overhead)?
- What errors or delays does the manual process cause, and what do those cost?
- How often do volume spikes create bottlenecks you can't handle?
If the answer to the first three bullets is “I don't know exactly,” that's normal, but it's also the first thing to fix. A rough estimate based on headcount and reported hours is enough to build on. You don't need precision; you need an order of magnitude.
Frame the return conservatively
CFOs are skeptical of AI projections because they've seen them oversell. The fastest way to lose the room is to project 90% automation savings on a first AI deployment. The fastest way to win it is to project 30% and explain exactly how you got there.
A credible projection has three components:
1. What changes, specifically
Not “it will automate the process”, but exactly what steps get automated, what human review remains, and what the new workflow looks like. If you can't describe the new workflow, you can't project the savings.
2. What you're assuming
Name your assumptions explicitly. “We're assuming 60% of tickets can be auto-routed without human review, based on our current ticket-type distribution.” Stating assumptions isn't a weakness, it shows you've thought it through and gives reviewers something specific to push back on rather than reject the whole thing.
3. What the downside looks like
Give a conservative scenario (40% automation instead of 60%) and show that it still works. If the business case only holds at the optimistic end, it's not a business case, it's a hope.
The Pilot is the argument, not the ask
The single most effective move in a business case for AI is to not ask for full implementation funding. Ask for Pilot funding instead.
A well-scoped AI Pilot, typically six to twelve weeks at a fixed price agreed up front, does two things. First, it produces real results on your real data, replacing projections with measurement. Second, it limits the downside: if the Pilot doesn't demonstrate the expected return, you've spent a small, known amount and learned something, not the cost of a full build and bet the business on it.
Most CFOs will approve a bounded experiment much more readily than an open-ended transformation. A Pilot framed correctly doesn't ask them to believe your projections. It asks them to fund the test that will produce the real numbers.
Address the “do nothing” cost
The status quo has a cost, and your proposal should name it. What happens to labor costs as volume grows if the manual process stays? What are competitors doing? What happens to error rates as headcount turns over and institutional knowledge erodes?
“Do nothing” isn't free. Making that visible, even in rough terms, reframes the decision from “should we spend money on AI” to “which path has the better expected outcome.”
What good looks like on one page
The strongest business cases are short. A one-page summary that covers: the problem and its current cost, the proposed solution and how it works, the return at conservative and base-case assumptions, the Pilot as a capped first step, and the cost of doing nothing, that's a decision your leadership team can actually make.
The longer the document, the harder the decision. If you can't fit the core argument on one page, you haven't finished thinking it through yet.