How to Build an AI Business Case Your CFO Will Actually Approve

The approval gap in AI is not a technology problem. It is a communication problem.

Organizations with genuinely valuable AI opportunities are failing to fund them, not because the value is not there, but because the proposal is built in the wrong language for the audience.

This guide walks through the five-step process for building a business case that survives the finance committee.

Step 1: Anchor to a Specific Business Problem

The most common mistake in AI proposals is starting with the technology. "We want to implement a generative AI system" is not a business case. It is a procurement request.

Start instead with a specific, measurable problem:

• "Our customer onboarding process takes 14 days and has a 23% dropout rate at the document verification stage."
• "Our sales team spends 40% of prospecting time on accounts that close below the median deal size."
• "Our accounts payable team processes 2,400 invoices per month with a 4.2% error rate requiring manual remediation."

Each of these statements contains a current state, a metric, and an implicit cost. That is where a business case begins.

Step 2: Quantify the Baseline Cost

Before you can claim a benefit, you need to know what the current state costs. This requires more precision than most proposals apply.

For process problems:

• Current headcount allocated to the process x fully-loaded cost per FTE
• Error rate x cost per error (remediation time, rework, customer churn)
• Process duration x opportunity cost (revenue deferred, capital tied up)

For revenue problems:

• Current conversion rate x pipeline volume x average deal size
• Customer acquisition cost in the current model
• Revenue leakage from missed cross-sell or upsell events

Do not estimate at the team level. Go to the individual task level. "The contracts review step takes 3 hours per deal on average, and we close 180 deals per year. At a $95/hour fully-loaded rate for a senior commercial counsel, that is $51,300 per year in legal time before error-handling costs."

That specificity earns credibility.

Step 3: Define the Future State Benefits

Benefits should be classified into three types:

Efficiency gains: The AI does the work faster or with less human input. Express as cost reduction or capacity reallocation, not headcount reduction unless you genuinely intend to reduce headcount. CFOs know the difference.

Quality improvements: The AI makes fewer errors. Express as avoided cost: fewer remediation events, lower return rates, reduced compliance exposure. Use your baseline error rate and cost per error from Step 2.

Revenue impact: The AI improves a conversion, retention, or expansion metric. This is the hardest to model credibly. Use conservative assumptions (10-20% of the theoretical maximum impact) and note your sources.

For each benefit, document the assumption, the data source supporting the assumption, and the person accountable for delivering the outcome. Anonymous benefits are not committed benefits.

Step 4: Build the Three-Scenario Model

Never present a single financial projection. CFOs are trained to distrust single-point estimates because they know the assumptions are optimistic.

Present three scenarios:

Conservative (30% probability weight): Implementation costs 20% higher than estimated. Time to value is 12 months. Year 1 benefits realize at 40% of projected. Adoption is slower than expected due to change management challenges.

Base case (50% probability weight): Costs meet estimate. Time to value is 8 months. Year 1 benefits realize at 70% of projected. Year 2 reaches full run rate.

Optimistic (20% probability weight): Implementation runs on time and on budget. Early adoption exceeds expectations. Year 1 benefits reach 90% of projected. Adjacent use cases emerge in Year 2.

Calculate NPV, IRR, and payback period for each scenario. Then calculate the probability-weighted NPV as your headline number: (0.3 x conservative NPV) + (0.5 x base NPV) + (0.2 x optimistic NPV).

Step 5: Frame the Cost of Inaction

Most business cases argue for the proposed investment. The strongest business cases also argue against the status quo.

Calculate the cost of a 12-month delay:

• Lost efficiency savings: 12 months x monthly net benefit in base scenario
• Competitive exposure: Estimate the revenue impact if a direct competitor deploys this capability 12 months ahead of you
• Talent signal cost: If AI deployment affects your employer brand with technical staff, model the premium you would pay to attract engineering talent in a non-AI-forward organization

Present this as a risk-adjusted "delay cost" alongside your investment returns. A business case that makes inaction look expensive is more persuasive than one that only makes action look attractive.

Presenting to the Finance Committee

Structure your presentation in 10 minutes or less:

• Minutes 1-2: The problem (specific, quantified, sourced)
• Minutes 3-4: The solution (what it does, not how it works)
• Minutes 5-7: The financial case (three-scenario model, headline numbers)
• Minutes 8-9: The risks and mitigations
• Minute 10: The specific decision requested

Bring the full model. Do not show the full model unless asked. The model is the backup, not the presentation.

The One Template You Need

The AI Business Case Playbook contains a complete financial model template pre-structured for this five-step approach, along with scenario worksheets, a cost taxonomy for common AI initiatives, and response scripts for the 10 most common CFO objections.

The approval gap is smaller than you think, if you speak the right language.