In short
An AI budget request tends to win approval when it reads like a finance document: one defined problem, a measured baseline, a full cost range, a small first stage and written stop criteria. Finance tends to decline requests for open-ended funding against benefits nobody can verify or date.
Key takeaways
- Finance usually rejects AI proposals it cannot test: indirect benefits, uncertain costs or no end date.
- An AI budget should be sized bottom-up for the first stage only, including integration, data, training and review.
- Written stop criteria build the executive buy in AI work needs, because finance can approve a bounded commitment.
Finance rarely rejects AI itself. It rejects proposals it cannot test. This guide, part of the AI implementation playbook, draws on the AI Strategy & Roadmap service and covers the objections finance raises, the answer to each, budget sizing, payback and a one-page proposal outline.
Why does finance reject AI proposals?
Finance rejects AI proposals when the value is indirect, the cost is uncertain and payback runs longer than ordinary technology spend. Each objection is a fair test, and a proposal that answers them in advance moves the discussion from whether to fund AI to how much to fund first.
- Unclear value. A July 2024 Gartner forecast expected at least 30% of generative AI projects to be dropped after the proof-of-concept stage by the end of 2025, with unclear business value among the named causes.
- Slow payback. Most of the 1,854 European and Middle Eastern executives in Deloitte's 2025 survey said a typical AI use case took two to four years to reach satisfactory ROI, against seven to 12 months for typical technology, per AWS figures Deloitte cites.
| Finance objection | Answer that gets a yes |
|---|---|
| Indirect benefit | Cost per unit of work, such as an invoice, against a measured baseline |
| Cost grows with usage | A cost range plus a trigger that pauses spend |
| Slow payback | A capped first stage with a dated payback assumption |
| Pilot fails on real data | A pilot on the business's own data |
| Nobody owns the outcome | A named business owner and a finance contact |
What does a CFO want to see in an AI business case?
A CFO wants a specific problem with a measured baseline, a full cost range, a dated payback assumption, named risks with owners and a small first stage with an exit. The business case for AI is judged like any capital request, with extra weight on what could go wrong.
Knowing how to build a business case for AI is mostly translation: finance reads benefits in dollars and hours, not model accuracy. A July 2025 Gartner webinar poll of 197 finance leaders, not a representative survey, found the focus moving to targeted use cases with measurable impact.
How do you show payback without made-up numbers?
Payback in months is one-time cost divided by monthly net benefit, where net benefit is units of work times time saved per unit times loaded hourly cost, minus run cost. Illustrative only: a hypothetical $60,000 cost and $5,000 monthly net benefit pay back in 12 months. For measuring it later, see how to prove AI ROI after a pilot.
How do you size an AI budget for a mid-sized business?
Size the AI budget bottom-up for the first stage only: one use case across build, data, licenses or usage, integration, training and review, shown as a range. A share of revenue or a competitor's AI investment says nothing about the use case being funded.
Usage pricing means spend rises with adoption. CloudZero, a cost-management vendor, advises ranges with written triggers, AI funding released in tranches tied to unit economics, and monthly reforecasting.
What stop-or-continue criteria reassure finance?
Knowing how to get budget approval for AI project work that is unproven comes down to starting small: fund one question, then tie later money to these rules.
- Continue if the measured gain beats normal variation and benefit exceeds run cost.
- Extend once, with a new date, if sample size or data problems blur the result.
- Stop if unit cost stays above the baseline or the gain vanishes on real data.
What does a one-page AI proposal that gets a yes look like?
A one-page AI proposal that gets a yes uses seven fixed headings in the order finance reads, two or three lines each, with detail in an appendix. The page then doubles as an AI proposal template and a standard technology business case for later requests.
- 01Problem and baseline: the process, unit of work and today's cost.
- 02Change: what AI does and what people review.
- 03Cost range: low, expected and high, one-time and recurring apart.
- 04Payback: formula, inputs and target date.
- 05Risks: data, adoption and usage cost, each owned.
- 06Ask: amount, scope and duration.
- 07Decision gate: date, criteria and decision maker.
Sarvadhi's AI Strategy & Roadmap service supplies the roadmap, budget assumptions, investment options and measurement plan behind this page; AI Proof of Concept & Pilot defines stage-one success metrics.
Conclusion
Finance tends to fund AI requests it can measure. A fundable AI budget names one problem, a cost range and a payback assumption, and asks for a small first stage with stop criteria. The business case for AI wins when finance can test it and exit it.
Run the 10-minute read before the next submission. To shape the request around where AI can create realistic value, Get an AI Readiness Assessment.
Filed under AI Strategy Consulting