Why Most AI Proposals Die in the Approval Meeting
If you're trying to figure out how to write an AI business case that actually gets approved, start with why most of them get rejected. It's almost never the technology. Our founder spent years on the other side of that table — as a CIO and operations director who approved some projects and killed plenty of others — and the pattern is consistent: proposals get funded when they're framed as money decisions and rejected when they're framed as technology decisions.
A CEO reviewing your proposal is holding it up against every other use of the same dollars: a new hire, more ad spend, a second location. "We should be using AI" loses to all of those, because all of those come with a number attached.
The three ways proposals fail
Nearly every rejected AI business case fails in one of three ways: there's no baseline (you can't say what the problem costs today), the benefits are adjectives instead of dollars ("faster," "streamlined," "more efficient"), or there's no answer to "what happens if it doesn't work?" Fix those three and you're ahead of 90% of the internal proposals your CEO has seen this year.
How to Write an AI Business Case: The 5-Part Structure
Keep the whole document to two or three pages. Executives don't approve 40-slide decks; they approve clear asks. Here's the structure we use with clients before any build begins.
Part 1: The problem, priced
One paragraph, ending in a dollar figure. Not "our intake process is slow" but "our two coordinators spend a combined 22 hours per week manually following up with leads, which costs $32,000 per year in loaded labor — and leads that wait more than an hour for a response close at roughly half the rate of leads contacted in five minutes."
Part 2: The proposed fix, scoped
Two or three sentences describing what gets built and — just as important — what doesn't. "An AI agent that responds to every inbound lead within two minutes, qualifies it, and books the appointment. It does not replace the coordinators; it removes the follow-up work from their week." Tight scope signals you know what you're doing.
Part 3: The math
All-in cost (build, software, maintenance), projected annual return, and payback period. More on how to make this credible below.
Part 4: The risks, pre-answered
Name the three most likely objections and your mitigation for each — before anyone raises them.
Part 5: The decision ask
Exactly what you want approved, by when, and the checkpoint where the project proves itself or gets killed. "Approve $24,000 for a 90-day implementation. At day 90 we review against the agreed metrics; if it misses, we shut it off."
1) The problem, priced in dollars. 2) The fix, tightly scoped. 3) The math: cost, return, payback period. 4) The risks, pre-answered. 5) The decision ask with a built-in kill checkpoint. Two to three pages, never forty slides.
Do the ROI Math Like a CFO, Not a Vendor
Vendor math assumes everything goes right. CFO math assumes half of it goes wrong and asks whether the project still clears. Build your case on CFO math: establish the baseline cost of the problem, estimate the improvement, then take a haircut.
A worked example
Take the lead follow-up agent above. Baseline: $32,000 per year in coordinator time, plus an estimated 15 recoverable lost jobs per year at $1,800 average margin — about $59,000 in total annual impact. Implementation: $24,000 build plus $500 per month in software, roughly $30,000 in year one. Now run three scenarios: optimistic (full $59,000 recovered), expected (70%, about $41,000), and conservative (40%, about $23,600). In the expected case, payback lands around month nine. In the conservative case, year one roughly breaks even and year two is clean profit, because build costs don't repeat.
Present all three scenarios and recommend approval only if the expected case pays back inside 12 months. That's the discipline behind our ROI-First Implementation Model — and if you want the full calculation method, we've written a step-by-step guide on calculating AI ROI before you spend a dollar.
If your business case only works in the optimistic scenario, you don't have a business case — you have a hope with a spreadsheet attached.
Answer the Risk Questions Before They're Asked
Every executive reviewing an AI proposal is silently asking four questions. Answer them in writing and you remove the easiest reasons to say no.
- "What if it doesn't work?" Define kill criteria up front: the specific metrics at day 60 and day 90 that trigger a shutdown. A proposal with a built-in exit is dramatically easier to approve than an open-ended commitment.
- "Who owns this?" Name one person accountable for the outcome — not "the team," not the vendor. Roughly 70% of AI projects fail, and missing ownership is one of the top operational causes.
- "What about our customers and data?" State where human review stays in the loop, what data the system touches, and what it's blocked from doing.
- "Why now?" Use your baseline: if the problem costs $59,000 per year, every month of waiting costs roughly $4,900. Waiting isn't neutral — it has a price.
Run a pre-mortem paragraph
Add three sentences: "The most likely way this project fails is X. We'll know by [date] because we're tracking Y. If that happens, we stop." Nothing builds executive confidence faster than a proposer who has already imagined the failure.
Day 60: agent handles at least 70% of inbound leads without human correction, or the build gets reworked at the vendor's expense. Day 90: booked appointments up at least 15% versus baseline, or the project shuts down and monthly software spend ends.
Size the Ask So "Yes" Is Easy
The final mistake is asking for too much at once. A $250,000 "AI transformation" requires faith. A $20,000–$40,000 first project with a 90-day checkpoint requires only a small, reversible decision — and its results fund the expansion conversation for you. (For realistic build budgets, see the real cost of building an AI agent in 2026.)
Structure the ask in phases: phase one is a single high-ROI workflow with agreed metrics; phase two expands only after phase one hits its numbers. When phase one is delivering measured returns, you'll stop pitching AI projects — leadership will start asking what's next.
If you'd rather pressure-test the numbers before putting your name on the document, that's exactly what we do. Book a free strategy call and we'll help you build the baseline, the scenarios, and the payback math for your highest-ROI workflow — before you spend a dollar on a build.