The AI business case template: building a board-ready investment proposal

Most AI proposals die in the boardroom for the same reason: they read like a vendor pitch, not an investment case. The numbers are rounded up, the risks are hand-waved, and the “return” is a single optimistic figure with no working shown. Boards have seen enough of these to be sceptical. If you want a yes, you need a credible ai business case template that treats directors as the intelligent, accountable adults they are.

The summary you can act on: a board-ready AI business case does four things — it quantifies value across four categories with an honesty discount applied, it counts the full cost (build, run, people and change), it shows payback and a three-year view, and it names the risks with the governance that contains them. Everything else is decoration.

This article gives you that structure. It is deliberately conservative, because the fastest way to lose credibility is to be caught rounding up. Use it alongside our AI ROI calculator to do the arithmetic, and treat the write-up below as the narrative wrapper around those numbers.

Start with the decision, not the technology

A board is not approving “AI”. It is approving a specific investment, for a specific outcome, with a specific owner. So the first page of any AI business case template should answer four questions in plain language:

• What are we proposing to do, in one or two sentences a non-technical director understands?
• What business outcome does it serve — cost, revenue, risk or capability?
• Who owns it and who is accountable if it underdelivers?
• What exactly are we asking the board to approve — budget, mandate, or a stage gate?

If you cannot state the ask in a sentence, you are not ready to present. Lead with the decision. The technology is an implementation detail that belongs later in the document.

The four value categories

Credible value cases separate the kinds of value, because each has a different level of certainty and a different owner. Lumping them into one headline number is the first credibility trap. Break the benefit into four categories.

1. Labour and cost efficiency. Time saved, work avoided, error rates reduced. This is the most common category and the most over-claimed. The trap is the “hours saved times hourly rate equals cash” fallacy — saving twenty minutes a day across a team rarely converts to a real headcount reduction or a real dollar. Boards know this. Claim efficiency only where the saving is harvestable: a role you genuinely won’t backfill, an outsourced volume you’ll stop paying for, a queue you can measurably shorten.

2. Revenue. Faster sales cycles, higher conversion, new products, better retention. This is the most attractive category and the hardest to defend, because revenue depends on the market, not just your tool. Be conservative and attribute carefully — if AI is one of five things influencing a conversion rate, don’t claim the whole lift.

3. Risk reduction. Fewer compliance breaches, better fraud detection, reduced key-person dependency, more consistent decisions. This is undervalued in most proposals because it is harder to put a number on. But for regulated AU organisations it is often the strongest argument. A defensible expected-loss reduction (probability times impact) can be more persuasive to an audit-and-risk committee than a speculative revenue line.

4. Strategic capability. Building the muscle — data foundations, governance, skills, reusable platforms — that makes the next ten use cases cheaper and faster. This is real value, but never dress it up as hard cash. Present it honestly as an option you are buying, and let the board weigh it as such.

For a deeper treatment of how a CFO stress-tests these numbers, our guide on measuring AI ROI is the companion piece to this one.

Apply the honesty discount

Here is the move that earns trust. After you have estimated the benefit in each category, apply a haircut to reflect uncertainty — and say so explicitly on the page. Claimed savings almost never land at 100 per cent, because adoption is partial, processes are messier than the pilot, and some of the “saved” time evaporates into other work.

A simple, defensible approach:

• Discount efficiency claims the hardest — often to 40 to 60 per cent of the theoretical figure — unless the saving is a genuinely harvestable cost.
• Discount revenue for attribution and ramp-up time.
• Risk reduction can hold up better if your probability and impact estimates are sober.
• Show both the gross and discounted benefit so directors see your working, not just your conclusion.

Presenting the haircut yourself is disarming. It signals you have thought like a sceptic before the board has to, and it is far stronger than being forced to concede the discount live in the room.

Count the full cost, not just the build

The second big credibility trap is understating cost by quoting only the build. Every serious AI business case template has four cost lines:

• Build — the one-off cost to design, integrate and deploy: internal effort, partner fees, data preparation, security review. Data readiness is routinely the biggest and most underestimated slice.
• Run — ongoing model, infrastructure, licensing and API consumption. Usage-based costs can scale in ways that surprise finance, so model a realistic volume, not a demo volume.
• Ops and people — the humans who keep it working: monitoring, prompt and model maintenance, human-in-the-loop review, retraining, support. AI is not a set-and-forget asset; it needs an owner and a budget line.
• Change — training, process redesign, communications, and the productivity dip while people learn. This is the cost most often left off, and the one that most often sinks the actual benefit.

If your run-and-maintain cost is a rounding error next to your build cost, you have almost certainly missed something. Our view on the honest cost of AI consulting in Australia unpacks where these numbers really land, and why the cheap-looking option often isn’t.

Payback and the three-year view

Boards think in time, not just totals. Give them two things.

Payback period. When does cumulative discounted benefit overtake cumulative cost? A payback measured in a small number of quarters is compelling; one measured in years needs a strong strategic-capability argument to carry it.

A three-year cash view. A simple table by year: costs, discounted benefits, net position, and cumulative net. Three years is usually the right horizon — long enough to capture the return, short enough that you are not fabricating precision about a fast-moving field. Include a net present value if your organisation uses one, and be explicit about the discount rate.

Two credibility notes. First, front-load the costs and back-load the benefits in your timing — that is how it actually happens, and a case that shows benefit from month one is not believable. Second, run a sensitivity view: what does the return look like if adoption is half what you hoped, or if run costs double? A case that survives a pessimistic scenario is far more persuasive than one that only works if everything goes right. The ROI calculator will do this arithmetic for you so you can present ranges, not a single fragile point estimate.

Risks and how they are governed

Do not hide the risks — govern them on the page. For an AU audience, a board will expect to see that you have thought about:

• Data privacy and security — how the initiative handles personal information consistent with the Privacy Act and the Australian Privacy Principles, and where data flows.
• Model risk — accuracy, bias, hallucination, and the controls (human review, evaluation, monitoring) that keep them in check.
• Regulatory and sector obligations — for financial services, expectations under frameworks such as APRA’s CPS 230 and CPS 234; and alignment with the Voluntary AI Safety Standard as a signal of responsible governance.
• Vendor and concentration risk — dependence on a single provider and your exit options.
• Delivery risk — the plausible ways the project underdelivers, and your stage gates.

The point is not to eliminate risk; it is to show each material risk has a named owner and a control. A short risk table — risk, likelihood, impact, mitigation, owner — does more for board confidence than three paragraphs of reassurance. If governance maturity is thin, say so and propose fixing it as part of the work, not after. Our AI strategy practice exists precisely to make this section defensible rather than aspirational.

The decision ask

Close where you opened: with a clean, specific ask. The strongest structure is staged funding tied to evidence. Rather than asking the board to commit the full multi-year spend on day one, ask for approval of a first stage — a pilot or a discovery — with a defined budget, defined success criteria, and a gate at which the board decides whether to proceed.

This de-risks the decision for directors and disciplines you. A short, focused engagement such as an AI readiness sprint is often the right first ask: it produces the evidence that turns an optimistic business case into a grounded one, before serious money is committed.

State plainly what you want approved, the amount, the owner, the next gate, and the date you will report back. Then stop.

A board-ready case is not the one with the biggest number. It is the one whose author has already argued against themselves and still believes it. Build it that way and you will spend the meeting discussing the decision, not defending your arithmetic.

If you want a second set of eyes on a proposal before it goes to your board — or help building the numbers that will withstand scrutiny — get in touch. We would rather help you present an honest case than a flattering one.