A board approves an AI business case when it asks for a specific decision about a specific problem. Name the workflow, show its baseline, cost the options, give a low and an expected return, set out the risks and controls, and ask for a first stage with a decision point. Ambition belongs in the roadmap, not the ask.
Key takeaways
- An AI business case asks for a decision, so it should fit on a few pages and end with a clear ask.
- According to Gartner, unclear business value, escalating costs and weak risk controls are among the reasons AI projects are abandoned.
- Present the return per workflow against a baseline, with a low case the board can rely on.
- Map risks to named controls: an accountable owner, data rules, human approval and staged access.
- Ask for a first stage with stated success and stop criteria, not the whole programme at once.
What is an AI business case?
An AI business case is a short document asking a board or leadership team to fund a defined piece of AI work. It states the problem, the options considered, the full cost, the expected return against a baseline, the risks and how they are controlled, and the measures and decision points that follow approval.
It is not a vision statement or a technology overview. Directors do not need to know how a model works. They need to know what will change in the business, what it costs, what could go wrong and when they will next get to decide.
Many boards are already asking. According to Microsoft and LinkedIn's 2024 Work Trend Index, 60% of leaders worry their company lacks a vision and plan for AI. A business case answers part of that worry with something concrete: one decision, properly argued.
Why do boards reject AI business cases?
Boards reject cases that promise transformation without a baseline, quote industry statistics instead of the business's own numbers, leave risk to a later phase, or ask for open-ended spend. Directors have read the headlines about failed AI projects. A case that ignores those failures invites the obvious questions.
The failure reasons are well documented, and boards know them. According to Gartner's 2024 forecast, at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. It cited poor data quality, inadequate risk controls, escalating costs or unclear business value. Each of those is a section a strong case answers before anyone asks.
Common weaknesses, and what fixes them:
| Weakness | Fix |
|---|---|
| "AI will transform the business" | One named workflow and the number it should move |
| Industry averages as evidence | The business's own baseline from system records |
| Licences as the only cost | Full cost: setup, review time, upkeep, outside help |
| Risk "to be addressed" | Named controls and an accountable owner |
| Whole programme in one request | A first stage with a decision date |
What should an AI business case include?
Seven sections cover it: the problem, the options, the cost, the return, the risks and controls, the measures and decision points, and the ask. Each can be a page or less. Together they let a board approve, amend or decline with confidence, which is the purpose of the document.
| Section | What it answers |
|---|---|
| Problem | Which workflow, what it costs today, why now |
| Options | Do nothing, switch on an existing tool, buy, or build |
| Cost | One-off and running costs, per option |
| Return | Low and expected cases, against a baseline, with payback |
| Risks and controls | What could go wrong and who stops it |
| Measures and decision points | What will be measured, by whom, and when the board decides again |
| The ask | The amount, the scope and the date of the next decision |
The options section deserves care. Including "do nothing" and "switch on what we already pay for" shows the board the case is honest. Often the cheapest option wins, and saying so builds credibility for the next request.
How do you present the return without overpromising?
Show the return per workflow, measured on a number the business already tracks, against a baseline from system records. Give a low case and an expected case, and calculate payback on the low one. Leave out self-reported hours and vendor percentages. A modest, defensible figure beats an impressive one the first audit overturns.
The wider evidence supports caution. According to McKinsey's 2025 State of AI survey, 88% of organisations use AI in at least one function. Yet only 39% report any earnings impact at the enterprise level. A board that knows this will trust a case that admits a range more than one that promises the average.
Use a value rule finance has agreed: cost per lead times volume, pipeline times win rate and margin, or contractor spend avoided. The guide to an AI ROI calculator sets out which inputs to include and which to leave out.
How do you address risk and governance in the case?
Map each risk to a control and an owner. Australia's National AI Centre sets out six essential practices for AI adoption, from deciding who is accountable to maintaining human control. Use them as the case's risk section, so directors see a recognised framework rather than a list of reassurances.
According to the National AI Centre's Guidance for AI Adoption, published in October 2025, the six practices are listed below. Each maps to something the case can commit to:
| Practice | What the case commits to |
|---|---|
| Decide who is accountable | A named owner for AI overall and for each workflow |
| Understand impacts and plan accordingly | Who the workflow affects, inside and outside the business |
| Measure and manage risks | Data rules: what may go into which tools |
| Share essential information | Telling staff and customers where AI is used |
| Test and monitor | A baseline, a test period and monthly review |
| Maintain human control | A named person approves anything sent, published or spent |
Access belongs here too. Commit that every connection to business systems starts read only, with write access granted one tool and one action at a time. Directors understand staged access, because it is how they already think about delegated authority.
How do you structure the ask so it gets approved?
Ask for a first stage, not the programme. Fund one or two workflows, or a short diagnostic that ranks the options, with success and stop criteria written down and a decision date about 90 days out. A board can approve a contained risk with a clear exit far more easily than an open commitment.
A staged ask has three parts:
- The first stage. Scope, cost and what will exist at the end.
- The criteria. The measure, the target against baseline, and the result that would stop the work.
- The next decision. When the board will see results and decide whether to fund the next stage.
That structure also sets up the reporting that follows. The guide to AI board reporting covers what to bring back.
Many cases stall because nobody has ranked the options in the first place. Skein's strategy diagnostic, which runs listen, map, rank and plan, produces those inputs: a ranked shortlist of three to five opportunities, a 90-day plan and a 12-month roadmap. The AI strategy page sets out the scope. That suits a leadership team that needs board approval and lacks a ranked, costed view. A business spending only on assistant seats needs a short policy, not a business case.
Frequently asked questions
What should an AI business case include?
It should include the problem and the workflow affected, the options considered, the full one-off and running cost, a low and expected return against a baseline, the risks with named controls, the measures and decision points after approval, and a clear ask for a first stage with a decision date.
How do you justify AI investment to a board?
Show one specific workflow, its current cost from system records, the expected change and its value, and a low case with payback. Address risk with named owners and controls. Ask for a contained first stage with success and stop criteria, so the board can approve a limited risk with a clear exit.
How long should an AI business case be?
A few pages is usually enough: roughly a page each for the problem, options, cost, return, risks, measures and the ask. Supporting detail, such as the cost model or baseline data, can sit in an appendix. Boards approve clear decisions faster than long documents.
What risks should an AI business case cover?
Cover who is accountable, what data may go into which tools, how output is checked before it is sent or published, how access to business systems is staged, how results will be tested and monitored, and what happens if the workflow underperforms. Map each risk to a named control and owner.
Should an AI business case include a do-nothing option?
Yes. A do-nothing option shows the cost of the current way of working, and a switch-on option shows what existing tools could achieve without new spend. Including both makes the case more credible, and sometimes the cheaper option is the right answer.
Related reading
- Microsoft and LinkedIn. "2024 Work Trend Index." Microsoft, 8 May 2024. https://blogs.microsoft.com/blog/2024/05/08/microsoft-and-linkedin-release-the-2024-work-trend-index-on-the-state-of-ai-at-work/
- Gartner. "Gartner predicts 30% of generative AI projects will be abandoned after proof of concept by end of 2025." Reported by Intelligent CIO, 5 August 2024. https://www.intelligentcio.com/apac/2024/08/05/gartner-predicts-30-of-generative-ai-projects-will-be-abandoned-after-proof-of-concept-by-end-of-2025/
- McKinsey. "The state of AI in 2025: Agents, innovation, and transformation." McKinsey, November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- Department of Industry, Science and Resources. "Guidance for AI Adoption." National AI Centre, 21 October 2025. https://www.industry.gov.au/publications/guidance-for-ai-adoption