The practical guide to assembling a defensible business case that incorporates economic analysis, total cost of ownership, risk assessment, and governance requirements: structured to survive whatever review and approval process your organisation uses.
They fail not because the opportunity is weak but because the document is. The typical AI agent business case is a technology pitch dressed up as a financial analysis: an impressive demo, a vendor quote, projected savings calculated by dividing headcount by automation percentage, and a timeline that assumes everything goes right.
This works for getting a pilot approved. It does not work for getting production funding, sustained budget allocation, or executive confidence that the investment is governed and the returns are real.
The business cases that succeed (the ones that survive finance review, risk committee scrutiny, and board-level questions) share a common structure. They address the business problem before the technology solution. They show fully-loaded economics including coordination tax, human transition costs, governance infrastructure, and failure reserves. They identify specific risks from a systematic catalogue rather than a generic “risks and mitigations” section. They connect to regulatory requirements the organisation actually faces. And they present returns as ranges with explicit assumptions rather than single-point projections.
This article provides that structure: section by section, with guidance on what each section must contain and why. A companion blank template is available at the end of this article. A future interactive tool, the Business Case Generator, will auto-populate the template using Corvair’s use case libraries, risk catalogue, and regulatory guidebook.
A complete AI agent business case contains ten sections. Not every organisation will use every section. Approval processes vary from a two-page executive summary at a startup to a sixty-page investment committee package at a global bank. The structure below is designed to be modular: use what your process requires, and know that each section exists because a real approval process somewhere has demanded it.
The single most important section. Many decision-makers will read only this page. It must stand alone.
What it contains:
Why it matters: The executive summary frames everything that follows. If it reads as a technology proposal, the business case will be evaluated as a technology expense. If it reads as a business investment with measurable returns and managed risks, it will be evaluated on its economics.
What it contains:
Common mistake: Describing the opportunity in generic terms (“AI will improve efficiency”). Every number should be traceable to a specific process metric that can be measured before and after deployment.
What it contains:
What it contains:
The nine-layer TCO framework from Article 2 is populated with specific estimates for this deployment:
Presentation format: Year-by-year table with low / expected / high estimates for each layer. Three-year and five-year totals where applicable, with cost-of-living adjustments on labour-dependent items.
The naive model comparison: Show what a traditional three-layer TCO model (build + infrastructure + consumption) would have estimated, and the percentage by which it understates true costs. This demonstrates analytical rigour and preemptively addresses “why is this more expensive than I expected?”
What it contains:
Three-year (or five-year) ROI calculation:
What it contains:
This is where most business cases are weakest. A generic “risks and mitigations” table with three rows does not constitute risk assessment. The business case should demonstrate that risk has been systematically identified and economically quantified.
What it contains:
Why this section matters: Regulatory alignment is not a compliance checkbox. It is a material economic variable. Deployments that satisfy regulatory requirements avoid remediation costs. Deployments that anticipate regulatory direction avoid future re-architecture costs. And in regulated industries, the ability to demonstrate governance to regulators is itself a competitive advantage.
What it contains:
What it contains:
What it contains:
Every organisation has its own approval process: investment committee packages, business case templates, stage-gate reviews, budget allocation cycles. The ten-section structure above is designed to be modular.
For lightweight approval processes (startup, small enterprise, single-decision-maker): Sections 1, 2, 3, 4, 5, and 10 are sufficient. Risk and regulatory sections can be abbreviated to one paragraph each within the executive summary.
For standard enterprise processes (mid-size financial services, corporate finance): All ten sections, with Sections 4 and 5 carrying the most weight. The TCO and value capture analysis are what finance teams scrutinise.
For heavy governance processes (global banks, public sector, prudentially regulated institutions): All ten sections at full depth, plus appendices for detailed risk scoring, regulatory mapping, and sensitivity analysis. Sections 6 and 7 (risk and regulatory) often receive as much scrutiny as the financial analysis.
For public sector (3+2 option year structure): Extend Sections 4 and 5 to include option-year projections with wider uncertainty ranges. Add a section on data sovereignty and deployment model constraints.
The key principle: the business case should match the rigour the organisation applies to equivalent investment decisions. An AI agent costing $500,000 over three years should receive the same analytical treatment as any other $500,000 investment. Not more (AI is not special), not less (AI is not exempt).
A companion interactive tool, the Business Case Generator, is under development. It will allow users to input their business problem, solution context, stakeholders, timeline, and budget parameters, and receive a pre-populated business case template drawing from:
The generated business case can be downloaded, edited, and submitted through the organisation’s approval process.
“We need help building the business case itself.” The Agentic AI Sprint Factory Sprint 0 phase produces the discovery artefacts: use case specification, data readiness assessment, architecture document, and acquisition strategy. These directly populate Sections 2, 3, and 8 of the business case. The subsequent sprint phases produce the DMAIC baseline metrics and Coordination Tax Impact Assessment that populate Sections 4 and 5 with measured data rather than estimates.
“We need the risk assessment and regulatory alignment sections to be credible.” The Agentic AI Risk & Controls Workshop produces an institution-specific risk taxonomy and controls checklist that populates Section 6. The MAS AIRG Readiness Assessment or AI Governance Framework Design produces the regulatory mapping and governance gap analysis for Section 7.
“We have multiple potential agent deployments and need to prioritise.” The AI Adoption Accelerator produces the tool landscape analysis, use case prioritisation, and adoption strategy that inform which business cases to build first. This ensures the portfolio-level economics described in Article 3 are considered alongside individual agent ROI.
Schedule a briefing to discuss your business case requirements.
A blank, downloadable business case template incorporating the ten-section structure described above is available as a companion resource:
Download the Agentic AI Business Case Template — see Article 7 of this series for the full annotated template.
This article is part of a seven-article series on the economics of agentic AI in financial services.
From discovery artefacts to regulatory mapping, Corvair services are designed to populate each section of your business case with measured data and credible analysis rather than estimates.
Schedule a Briefing Explore the Sprint Factory