Most AI budget requests get declined for the same reason: they ask to buy something before anyone can see what's happening. A purchase with no baseline behind it lands as a cost line with no owner, and that defers to next quarter without an argument.
Across the companies we've asked about this, the funded asks ran the other way around. They started with visibility, which is cheap and fast, and let what they saw decide what to buy. Get the data first, and the data picks the spend.
This is written for the person who already owns AI adoption and now needs money behind it. Here's the pitch, in the order that works, and where each part usually breaks.
1. Open with the number you're moving
Name a metric your CFO already tracks. Pipeline velocity stalls at one stage, and that stage carries a number.
Support tracks cost per ticket for its highest-volume type. Onboarding tracks days-to-first-value, and renewals carry a rate in one segment.
Lead with the metric and the spend becomes an implementation detail. Lead with the purchase and you've framed your own work as shopping.
If you can't name the number yet, hold the pitch. That takes about two weeks to fix.
2. Show what you can already see
Bring what's true today. How many seats you bought, how many people opened them, where your chosen metric sits right now, and how far your best performers sit from your median on it.
That last gap moves the room. Your top quartile already closes faster, or resolves tickets cheaper, and the distance between them and everyone else is money sitting on the table this quarter. Getting a clear look at that gap costs a fraction of anything you'd buy to close it.
Bring the unflattering half yourself. A CFO who finds it later stops trusting the deck, and one who hears it from you starts treating you as the person who knows.
3. Name the mechanism, in one sentence
Name a mechanism, not a category. Give a skeptic something to picture: the three reps who close fastest prepare for renewal calls a particular way, and we're going to make that the way everyone does it.
A mechanism survives the first clarifying question. A category collects one, then gets deferred.
4. Size the ask against the status quo
Most owners undersell here. Your ask doesn't compete with zero, it competes with what today already costs, and today carries a price.
Your AI spend already runs on the card. Your team already spends hours rebuilding the same context by hand. And a senior person keeps losing days to answering "what's actually happening," which costs real money whether or not anyone books it that way.
Ask for a quarter, not a transformation. Name a number, a date, and a decision at the end of it.
5. Give the decision an exit
Say out loud what you'll do if it doesn't work. "If the metric hasn't moved by day 90, we stop, and I'll tell you why."
Naming the day you'd walk away is worth more than any projection you could put beside it. It also turns a permanent budget line into a bounded experiment, which is a much easier yes.
6. Close on what you'll report, and when
Commit to the readout: the baseline, the change, the recommendation. Put a date on it, then hit that date even if the news is mixed. The second ask is decided by how the first report landed.
The outline, one slide per part
| Slide | What's on it | Where it usually breaks |
|---|---|---|
| 1 · The number | One metric your CFO already tracks, and where it sits today | Naming a purchase instead of a metric |
| 2 · What you can see | Seats bought vs. opened, the metric now, best-vs-median gap | Hiding the unflattering half |
| 3 · The mechanism | One sentence a skeptic can picture | A category instead of a mechanism |
| 4 · The ask | Visibility first, then the spend the data justifies | Competing against zero instead of today's cost |
| 5 · The exit | What you stop doing if the number hasn't moved by day 90 | No exit, so it reads as permanent |
| 6 · The report | Baseline, change, recommendation, on a fixed date | No date, or a date you then miss |
What changes after the first yes
The second pitch goes easier, and not because you got better at pitching. You now have a number with history behind it, which you lacked the first time.
That's also when this stops being a slide problem. Reporting a metric monthly means somebody keeps producing it, and by the third month "who pulls this number" is a real question. Better answered on purpose than by default.
So: pick the number, get a clear look at where it sits, name one mechanism, and put a date on the readout. Then go make the ask.