ctaio.dev Ask AI Subscribe free

AI ROI / AI Business Case

AI ROI · Template

The AI Business Case

A Board-Ready Template for 2026

Most AI business cases die in finance review for the same reason: they show only the upside, ignore adoption and capture haircuts, and don\u2019t set kill criteria. This template gives you the nine sections every defensible AI business case has, the build-vs-buy decision matrix CTOs and CAIOs actually use, and the ten questions every CFO will ask. Use it before funding, after funding, and at the quarterly review.

30-SECOND EXECUTIVE TAKEAWAY

  • Co-author with finance. Business cases written by tech alone inflate value; written by business alone underestimate cost. Both signatures or neither.
  • Set kill criteria up front. The 95% AI failure rate is partly a function of business cases that have no exit clause.
  • Five to ten pages, not fifty. Length that exceeds the page limit usually hides weak assumptions in volume.

The structure that survives review

Most failed AI business cases share a structural defect, not a math defect. They lead with the AI tool instead of the problem. They show the upside without the haircuts. They cite vendor case studies as if they predict outcomes in a different organization. They don\u2019t document what would cause us to stop. They put the build-vs-buy decision into the appendix when it should be in the executive summary.

The nine sections below are what a defensible AI business case looks like in 2026. The structure works for productivity tooling rollouts, customer-facing AI features, agentic deployments, and platform investments. Adapt the depth per section to the size of the investment, but keep the structure intact. The order matters; finance reviews the early sections most carefully.

NINE SECTIONS

The standard AI business case structure

Each section is one paragraph to one page in the final document. Use the AI ROI calculator for section 4 and the AI risk management guide for section 6.

01

Problem & current cost

The specific business problem in measurable terms (e.g., support handle time, content production cost, sales cycle length). The current cost of not solving it.

02

Proposed AI solution

The specific solution. Foundation model, deployment pattern (RAG, agent, fine-tuned), integrations. One paragraph; this isn’t the architecture document.

03

Build vs buy decision

The choice and the reasoning. Off-the-shelf vendor, build on a foundation model, or hybrid. Decision matrix below.

04

Realistic ROI math

Annual benefit and cost with explicit adoption rate and productivity capture rate. Use the AI ROI calculator and document the inputs.

05

Total cost of ownership (3 years)

Tool cost, inference at production scale, integration, governance, change management. Year 1 is highest. Year 3 baseline is steady-state.

06

Risk register entries

Top 5–10 risks specific to this AI program. Includes prompt injection, accuracy mismatch, vendor lock-in, regulatory exposure. Each with owner and mitigation.

07

Success metrics

How we measure success at 3 months, 6 months, 12 months. Productivity, financial, adoption, accuracy.

08

Kill criteria

The specific signals that trigger restructure or termination. Set up front. Reviewed quarterly.

09

Governance & ownership

Business sponsor, technical sponsor, finance co-author, AI risk register entry, review cadence.

BUILD VS BUY

The build-vs-buy decision matrix

The build-vs-buy decision belongs in section 3 of the business case and in the executive summary. Use the matrix below to test the choice; most enterprises buy more often than they think they should and build less often than the technical team would prefer.

ScenarioRecommendationReason
Common business problem, mature SaaS option exists Buy Faster time-to-value, lower TCO, vendor handles model upgrades and security
Differentiated use case, accuracy or domain-specific knowledge required Build (on foundation model) Off-the-shelf can’t hit your accuracy or domain-specificity requirement
Highly regulated industry, audit-grade evidence required Build or specialized vendor Generic SaaS often can’t meet sector-specific compliance; specialized vendors may exist
Need to integrate deeply with proprietary internal systems Build (with vendor support) Integration cost is the dominant TCO line; building gives flexibility
Productivity tooling for general knowledge workers Buy (M365 Copilot, ChatGPT Enterprise, etc.) No differentiation from building; vendor scale gives better economics
No ML engineering muscle in-house Buy Build creates ongoing maintenance burden the org isn’t resourced for; technical debt accumulates
Use case is core to the product or revenue model Build or strategic partnership Outsourcing the core competency leaves long-term strategic exposure

CFO REVIEW

The 10 questions every CFO will ask

Pre-answer these in the business case before the meeting. A business case that doesn\u2019t address all ten won\u2019t survive serious finance review, and the questions that aren\u2019t addressed are the ones that get asked first.

  1. What’s the adoption rate assumption, and what data supports it?
  2. How does the inference cost scale from pilot to production traffic?
  3. What’s the productivity capture rate, and how was it calibrated?
  4. What’s the kill criteria, and who reviews it?
  5. What’s the contingency if the foundation model vendor changes pricing?
  6. How does this program fit alongside our other AI investments?
  7. What’s the regulatory exposure, and who owns compliance?
  8. What’s the change management investment, and how is it tracked?
  9. What’s the post-mortem trigger if this program underperforms?
  10. What’s the comparable benchmark from other organizations that have done this?

Run your numbers through the AI ROI calculator first. The defaults applied there are roughly the haircuts a thoughtful CFO will apply mentally during the review.

AI Business Case: Frequently Asked Questions

What goes in an AI business case?
A defensible AI business case has nine sections: the problem and current cost, the proposed AI solution, the build-vs-buy decision, the realistic ROI math (with adoption and capture haircuts), the total cost of ownership over three years, the risk register entries, the success metrics, the kill criteria, and the governance ownership. Most failing AI business cases skip the haircuts, the kill criteria, or both. The result is a pretty number on the cover page that doesn’t survive the first executive review.
How is an AI business case different from a software business case?
Two material differences. First, AI cost models are non-linear: pilot economics rarely match production economics because inference cost grows with usage and context-window size. Second, AI value capture is uncertain: productivity gains rarely convert to financial value 1:1 because the saved hours often don’t become recovered hours. The business case structure must explicitly account for both, or the number is not defensible. See our AI ROI hub for the cost-category framework.
Build vs buy: how do I decide?
Three questions. (1) Is your use case differentiated, or is the problem common across enterprises? Common problems should buy; differentiated problems may justify build. (2) Do you have ML and AI engineering muscle in-house, including the maintenance capacity for prompts and models post-launch? Without it, build becomes ongoing technical debt. (3) Does the cost-of-being-wrong require accuracy or governance that no off-the-shelf vendor can prove? If yes, build may be the only option. The decision matrix below lays out the trade-offs by scenario.
What’s the typical ROI horizon for an AI business case?
Aim for payback inside 12 months for productivity-style use cases, inside 18 months for transformational use cases, and never more than 24 months because the foundation model market changes too fast for longer assumptions to hold. The AI ROI calculator at ctaio.dev/en/ai-roi/calculator/ applies field-tested haircuts; if the calculator shows payback above 24 months even after generous adjustments, the business case is fragile and probably won’t be funded by a serious finance review.
What kill criteria belong in an AI business case?
Four signals, set up front. Time-to-value at full scale slipping more than 6 months at a time. Adoption plateaued more than 50% below business case assumption. Operating cost more than 3x pilot model. Required vs achievable accuracy gap wider than human-review cost can close. Any one is a yellow flag; two means restructure or kill. See our AI project failure rate guide for the full kill-criteria framework.
Who should own the AI business case?
A named business sponsor (the executive who owns the outcome the AI is supposed to improve), with the CAIO or CTO as technical sponsor and finance as a co-author. Two common failure modes: a technical leader writes the business case alone and inflates the value capture, or a business sponsor writes it alone and underestimates the cost. Co-authoring catches both. The business case should be reviewed by the CFO’s office before it goes to the funding committee.
How long should the business case document be?
Five to ten pages. The board version is one page summary. Anything longer signals that the case isn’t actually clear, or that the team is hiding weak assumptions in volume. Use the structure below as the table of contents; a section that takes more than a page is probably the section to scrutinize.
·
Thomas Prommer
Thomas Prommer Technology Executive — CTO/CIO/CTAIO

These salary reports are built on firsthand hiring experience across 20+ years of engineering leadership (adidas, $9B platform, 500+ engineers) and a proprietary network of 200+ executive recruiters and headhunters who share placement data with us directly. As a top-1% expert on institutional investor networks, I've conducted 200+ technical due diligence consultations for PE/VC firms including Blackstone, Bain Capital, and Berenberg — work that requires current, accurate compensation benchmarks across every seniority level. Our team cross-references recruiter data with BLS statistics, job board salary disclosures, and executive compensation surveys to produce ranges you can actually negotiate with.

Continue the AI ROI cluster

A defensible business case is the start. The calculator hardens the math; the failure-rate guide sets the kill criteria.