How to Build an AI Business Case Executives Will Approve

Most AI business cases fail the CFO test not because the opportunity is wrong, but because the cost model is incomplete and the timeline to value is wishful. Here is the framework that changes that.

Why AI Business Cases Fail the CFO Test

The typical AI business case has three elements: a large potential productivity gain, a software license cost, and an implied ROI. What it omits is everything the finance team will ask about. Change management. Data readiness. Integration work. Ongoing operations. The timeline from pilot completion to the state where the productivity gain is actually being realized at scale.

The McKinsey Global Institute's June 2023 analysis of generative AI's economic potential estimated that the technology could deliver substantial productivity value across industries. That framing is useful for setting strategic context with a board. It is not a business case for a specific deployment. Your business case needs to be grounded in the specific process you are changing, the specific cost structure of that process today, and the specific changes the AI will produce, measured and sourced.

The ROI Formula for AI Initiatives

AI ROI follows the same logic as any capital investment: ROI = (Value Created - Total Cost of Ownership) / Time to Value. Each term requires careful definition.

Value Created comes from three sources: labor time saved (hours reduced multiplied by fully-loaded cost per hour), quality improvement (error reduction or faster cycle time, valued in downstream cost or revenue impact), and new capability (tasks the organization could not do before, valued at what you would have paid to produce equivalent output). Every value claim needs a measurement method and a baseline. "We expect to save 20% of contract review time" is a claim. "Legal reviews an average of 180 contracts per month, each taking 3.4 hours at a fully-loaded cost of $210 per hour, and we will validate whether AI reduces review time by 20% in a 90-day pilot" is a business case.

Total Cost of Ownership has five components. Software and compute are visible and well-scoped. Data engineering, the work to prepare and maintain the data the AI system needs, is the most consistently underestimated cost category. Integration, change management, and ongoing operations round out the five. Projects that capture only the first component and ignore the other four produce timelines and budgets that collapse during implementation.

Time to Value is the elapsed time from project start to the state where value is actually being realized at scale. Optimistic time-to-value assumptions are the most common reason AI business cases produce disappointment. Break it into milestones: pilot running, pilot validated on real process, full rollout to first department, full rollout complete. Present the ROI at each milestone separately, and let the CFO see what the return looks like if the timeline extends by three months or six months.

The Evidence on Knowledge Worker Productivity

Research from BCG and MIT Sloan Management School (Dell'Acqua et al., 2023) is frequently cited in AI business cases. The study found that consultants using AI on a defined set of creative and analytical tasks outperformed the control group by 12.2% on average. That is a real and meaningful number, and it comes from a well-designed experiment.

The same study found that for tasks outside the AI's competence, participants using AI performed worse than those who did not, because they over-trusted AI outputs they could not evaluate. The implication for business case construction is precise: cite the 12.2% finding only for use cases where the target task is similar to the study conditions and where your team has a way to verify AI output quality. For tasks where output quality is hard to evaluate, the business case needs to account for the oversight cost and the risk that AI-assisted work requires more correction than unassisted work.

The Three Failure Modes to Name in Your Business Case

A business case that names its own failure modes is more credible, not less. CFOs have seen too many AI proposals that arrived without acknowledgment of risk. Naming the three most common failure modes and explaining how your plan addresses them signals that the proposal has been thought through.

Over-scoping the MVP

The most common failure mode is designing a first deployment that tries to solve the entire problem rather than one well-defined slice. A scoped pilot produces clear signal about whether the approach works. An over-scoped pilot produces ambiguous results that make it impossible to learn anything. Your business case should specify what is in scope for the pilot and what is explicitly deferred to Phase 2.

Ignoring the Change Management Budget

The technology procurement decision and the organizational change decision are different decisions, and most AI business cases only address the first. Change management includes training, workflow redesign, and the productivity dip during transition. Organizations that treat AI deployment as a technology project consistently see weaker realized value than the business case projected because adoption is lower than assumed.

Underestimating Data Readiness Costs

The assumption that "our data is in good shape" almost always collapses when the team touches the data in earnest. Data preparation costs appear in two forms: the upfront work to make data usable for the pilot, and the ongoing maintenance required to keep it usable as the deployment expands. The business case should include a data readiness estimate from the engineer who will actually do the work, not from the vendor who wants to start the project.

For the full framework, including a TCO estimator and the complete ROI model, see Module 2 of the AI for C-Suite Leaders course.

What to Put in Front of the CFO

The CFO will approve an AI investment when four questions have clear answers: what is the current baseline, what will change after deployment, how will the change be measured, and who owns the measurement? A business case that cannot answer all four is not ready to present. A business case that answers all four, with sourced numbers and honest timelines, has a high probability of approval, because it demonstrates the rigor that distinguishes an investment from an experiment.

FrugalGPT research (Chen, Zaharia, Zou, arXiv:2310.11409) demonstrates that inference cost optimization, routing queries to different model tiers based on complexity, can reduce AI operating costs materially. If your TCO calculation includes ongoing API costs at scale, include a note on whether your architecture accounts for inference cost optimization, or whether that is a Phase 3 initiative.