Building an AI ROI Case Your CEO Will Actually Take Seriously

Building an AI business case is a finance skill, full stop. If you work with clients on this, you’ve probably seen the same pattern I have: a client has a thousand places they’d like to deploy AI, and the moment you actually dig into it, the ROI doesn’t hold up — the task in question might take hours or days to properly build out, to save a few minutes a month. Figuring out where the real ROI is, and where it just isn’t there yet, is the actual work.

Most AI pitches fail because the case never actually got built. Right now, a lot of spending is sliding through simply because the C-suite likes being able to say they’re investing in AI. Boards like that story. Investors like that story, especially at public companies. So the spending happens easily. What’s changing is that people are starting to ask the follow-up question: we’re spending a lot on this — can you point to the return? And that’s a much harder question to answer than it sounds, partly because some of what AI is doing might just be keeping you competitive, which is nearly impossible to measure cleanly. If your spend goes up and your revenue and EBITDA stay flat, is that AI failing to deliver, or is that AI keeping you exactly where you’d have fallen behind without it? Genuinely hard to untangle.

The fix is structuring an AI case exactly like any other capital or procurement case. On the cost side: subscription or token cost, sure, but also implementation time and — the part people consistently underestimate — data cleanup and change management. A lot of AI tools are technically ready to turn on immediately. In practice, most clients I’ve worked with have months of data strategy work before the tool is actually usable, because the underlying data was never captured cleanly enough to build on. That’s real implementation time, and it belongs in the cost side of your case.

On the return side, it’s genuinely harder to measure. Hours saved, translated into dollars. Error reduction, translated into avoided cost. Revenue or capacity impact, if it’s real. I’ll be honest — if I tried to calculate my own personal ROI from AI, I couldn’t give you a clean number. Everything I’ve built since leaving PwC has had AI involved somewhere, so I don’t have a clean “before” to compare against. That’s a common problem, and it’s worth naming rather than papering over with a false-precision figure.

A few things that actually move the needle when you present this. First, use your company’s own numbers for what an hour is worth — someone in FP&A almost certainly already has that figure, so don’t invent your own. When someone questions your math, you want to be able to say the number came from the same place every other business case’s numbers come from. Second, acknowledge forecast uncertainty explicitly rather than presenting one falsely confident figure. State what could go wrong, what could change — a better model could ship next month and change your whole calculation. That’s not a weakness in your case. It’s what every other capital case does too, and pretending otherwise reads as less credible, not more. Third, and this is the framing that actually gets you in the door: the CEO-level pitch is a cost-and-growth story first, an AI story second. AI is what gets you a hearing. It shouldn’t be the thing you’re selling.

And once it’s approved, track your actual results against the case you made. That follow-through is what earns you a faster, less-scrutinized approval the next time you come back asking for budget.

Key takeaways

• Build the AI case exactly like a capital investment case: cost side, return side, payback period.
• Cost includes implementation and change management time, not just subscription or token cost — that’s usually the part people underestimate.
• Translate soft claims like “saves time” into a specific dollar figure, using numbers your own FP&A team already tracks.
• Acknowledge forecast uncertainty explicitly rather than presenting a single falsely precise number.
• Frame the pitch as a cost-and-growth story first, an AI story second — and track actual results against the case after approval.

Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: Building an AI ROI Case Your CEO Will Actually Take Seriously