AI Business Case Template That Gets Past the CFO

Most AI investment proposals fail the first time they reach the CFO. Not because AI does not deliver value, but because the business case is built in a way that invites scepticism. Costs are underestimated by 40 to 60 percent. Benefits are overstated or unattributable. The financial model has no downside scenario. And the payback assumption is based on a vendor's marketing deck rather than your own data.

After reviewing investment proposals at more than 200 enterprises and advising on over $4.2 billion in AI investment approvals, we have identified the structural patterns that separate approved cases from rejected ones. This article gives you the complete template.

Why AI Business Cases Get Rejected

Before presenting the template, it is worth being specific about the failure modes. In our experience advising on AI investment governance across Fortune 500 companies, rejections cluster around five structural problems.

Incomplete cost taxonomy. The initial cost estimate includes compute and software licences, but omits data engineering (typically 3 to 4x the model development cost), change management, integration work, ongoing monitoring, and retraining cycles. When the actual costs surface during implementation, the case retroactively looks dishonest.

Single-scenario modelling. Presenting one ROI number without a conservative, base, and optimistic scenario signals that the author has not stress-tested the assumptions. CFOs expect to see what happens if adoption is 40 percent lower than planned, or if the model requires a 6-month retraining cycle.

Unattributable benefit claims. "AI will improve customer satisfaction" does not pass CFO scrutiny. Benefits must be specific, attributable, and measurable. "Reducing call deflection by 18 percent at current contact centre volume generates $3.4M in annual savings, verifiable through call volume data already tracked in the CRM" is a defensible claim.

No governance cost. EU AI Act compliance, model risk management, and audit documentation are not free. Excluding governance from the cost model means the case gets sent back when the legal or compliance team asks the obvious question.

Benefit realisation timeline mismatch. A case that shows benefits starting in month 2 and reaching full run-rate by month 6 will be challenged by any CFO who has seen an enterprise software rollout. Production deployment takes longer than optimists plan. Build the timeline from actual project data, not best-case assumptions.

The Seven-Section Business Case Template

This structure has been refined across more than 200 investment approvals. Each section addresses the specific questions a CFO or investment committee will ask.

• Executive Summary (1 page)

  The decision, the investment required, the expected return across three scenarios, and the consequences of not acting. CFOs read this first and often only. Lead with the financial outcome, not the AI technology.

• Problem Statement and Strategic Context

  What specific, quantified problem is this solving? How large is it? What is the cost of the status quo? What evidence exists that the problem is solvable with AI? Anchor to data already in the business, not to generic AI potential.

• Proposed Solution and Scope Boundary

  What exactly is being built and deployed? What is explicitly out of scope? A well-defined scope boundary prevents scope creep and sets realistic expectations. Include a one-paragraph description of the technical approach accessible to a non-technical reader.

• Complete Investment Requirement

  All costs across four categories: build costs (data engineering, model development, integration, testing), operating costs (compute, licences, monitoring, retraining), governance costs (compliance, documentation, model risk review), and change management costs (training, process redesign, communications). See cost taxonomy below.

• Three-Scenario Benefit Analysis

  Conservative (40 to 50 percent of planned adoption, 6 months to full run-rate), base (planned adoption, 3 months to run-rate), optimistic (110 percent of planned adoption, 2 months to run-rate). For each scenario: gross benefit, net benefit, ROI, and payback period. Include the attribution methodology for each benefit claim.

• Risk Register and Mitigations

  Technical risks (data quality, model performance), adoption risks (change resistance, process integration), regulatory risks (EU AI Act, sector-specific requirements), and vendor risks (dependency, lock-in, pricing changes). For each risk: probability, financial impact, and specific mitigation already planned.

• Governance and Measurement Framework

  How will you know whether the investment is delivering? What are the leading indicators (model performance, adoption rate) and lagging indicators (cost savings, revenue impact)? What are the stage gates? What triggers a pivot or stop decision?

The Complete AI Cost Taxonomy

The most common mistake in AI business cases is the cost taxonomy. Enterprises consistently underestimate total cost of ownership by 40 to 60 percent because they only capture the visible costs. Here is the complete model.

Three-Scenario Financial Model

Present three scenarios, not one. The conservative scenario is not pessimism; it is intellectual honesty that builds CFO confidence. A CFO who sees that the investment is still positive in the conservative scenario is far more likely to approve it than one who sees a single optimistic number.

The numbers above are illustrative, calibrated to our benchmark of 340% average 3-year ROI across 200+ enterprise deployments. Your actual scenario modelling should be grounded in the specific assumptions for your use case, your organisation's adoption history, and your cost actuals from comparable projects.

Benefit Attribution: Making Claims Defensible

Every benefit claim in the business case needs a clear attribution chain. The chain runs: AI model output → process change → measurable business outcome → financial impact. Any break in this chain creates a gap that a CFO will find and challenge.

For a claims automation use case, the chain looks like this: "The model classifies 89 percent of claims as straight-through eligible (AI output) → those claims are processed without human review (process change) → average processing time falls from 4.2 days to 0.8 days for 89 percent of volume (measurable outcome) → with $42 of adjuster cost per manual claim and 2.1M claims annually, that represents $67M in annualised savings, reduced by the 40 percent of claims still requiring human review" (financial impact). That is a defensible claim. "AI will reduce claims costs significantly" is not.

The Five CFO Questions You Must Answer

These are the five questions that most frequently cause business cases to get sent back from the CFO's office. Answer them explicitly in your proposal, do not wait to be asked.

• "How do we know your benefit estimates are credible?"
  Show the attribution chain for each benefit claim. Reference internal data sources (CRM records, cost accounting systems) rather than external benchmarks. Where you use industry benchmarks, apply a conservative discount (40 to 50 percent) and explain why your organisation might perform differently.
• "What happens if the model does not perform as expected?"
  Show the conservative scenario. Describe the fallback process (does the existing workflow continue in parallel?). Specify the performance threshold below which you would pause deployment and the process for deciding whether to remediate or stop.
• "How is this different from every other AI initiative that has not delivered?"
  Be specific about what went wrong previously. If prior AI investments underperformed, address that directly: what has changed in readiness, governance, or approach? If this is the first AI investment, address the readiness question directly rather than assuming the CFO is not thinking about it.
• "What are the regulatory and compliance costs?"
  Explicitly include EU AI Act risk classification, model risk management requirements (SR 11-7 for financial services), GDPR implications for training data, and any sector-specific regulatory requirements. Show the legal or compliance team has reviewed the proposal.
• "What is the exit strategy if this does not work out?"
  Specify the conditions under which investment would be paused or stopped, and what the exit cost would be (data cleanup, vendor contract termination, process reversion). A CFO who understands the downside is far more comfortable approving the upside.

Governance Framing for the Investment Committee

Investment committees increasingly expect AI proposals to address governance explicitly. The EU AI Act has made this a legal requirement for many high-risk use cases, not just a best practice. Your business case should include a governance appendix covering risk classification (which tier under the EU AI Act), model documentation standards, oversight mechanisms during operation, and the process for handling model failures or incidents.

Proposals that address governance proactively move faster through approval cycles. Proposals that leave it to the legal team to raise during due diligence get delayed by 4 to 8 weeks on average.

Before You Submit: A 10-Point Checklist

Before sending your business case to the CFO or investment committee, verify each of the following. If any item is incomplete, fix it before submission.

1. All benefit claims have an explicit attribution chain to measurable business data.
2. The cost model includes all seven cost categories in the taxonomy above.
3. Three financial scenarios (conservative, base, optimistic) are presented with different adoption and ramp assumptions.
4. The conservative scenario still shows positive ROI within 36 months.
5. Governance costs are explicitly included (compliance review, documentation, model risk management).
6. The EU AI Act risk classification is addressed.
7. A risk register with financial impact estimates is included.
8. Stage gates are defined with specific decision criteria (not "review progress").
9. The exit strategy is specified with an estimated exit cost.
10. The case has been reviewed by the legal or compliance team before submission.

A business case that passes this checklist will not guarantee approval. But it will demonstrate the rigour that separates a credible proposal from a wishful one, and it will materially reduce the probability of being sent back for revision.

For organisations that want independent support in building or reviewing their AI business case, our AI Strategy team works alongside internal finance and technology teams to build investment cases that survive CFO scrutiny. We have supported over $4.2 billion in AI investment approvals across financial services, healthcare, manufacturing, and retail. The free AI assessment is the fastest way to understand where you stand before committing to a major investment.