How to Write a Business Case for AI Spend Your Boss or Partners Will Back
2026-11-10
Asking for money for AI works best when you treat it like any other business decision. Your boss or partners want to know what problem you are solving, what it costs, what it returns and what could go wrong. This guide walks through each part of a one-page case. You end with a document you can hand over and a small first step that is easy to approve. The government's Guidance for AI Adoption can back up the risk and governance sections.
Step 1: Start with a problem they already feel
Open with a problem in the business, not with AI. "We lose enquiries because replies take a day" lands better than "AI could transform our workflow".
Pick a problem that your decision makers have complained about, or that shows up in the numbers. Write it in two sentences with one concrete example. Use facts from your own business: how many quotes sat unanswered last month, how many hours the office manager spends re-keying invoices. If you are unsure where to start, read is my business ready for AI first.
Step 2: Gather your own numbers
Avoid industry statistics and vendor claims. They are easy to challenge and may not apply to a business your size. Use what you can verify:
- Hours per week the task takes, from a quick time sample.
- Number of times it happens.
- Who does it and roughly what their time costs the business.
- Errors, rework or missed items you can count.
- Lost or slow outcomes you can point to, such as unanswered enquiries.
If you do not have data, collect it for two weeks before you write the case. The data is the strongest part of your argument, and a baseline lets you check the result later. See how to measure whether an AI change is working.
Step 3: Describe the proposal in plain words
Say what you want to do, who it affects and what it will replace. One short paragraph. For example: "Use ChatGPT's business plan for the office team to draft customer replies and summarise supplier emails, with a manager reviewing anything customer-facing."
Avoid jargon. If a partner has to ask what a term means, rewrite the sentence.
Step 4: Show the options
Decision makers like to see that you considered alternatives. A short table works well.
| Option | What it involves | Upside | Downside |
|---|---|---|---|
| Do nothing | Keep the current process | No cost, no disruption | The problem continues |
| Use what we already have | Switch on features in current software | Little extra cost | May not cover the need |
| Small trial of a new tool or automation | Time-boxed, one process | Low risk, clear evidence | Takes some staff time |
| Larger rollout | Several processes, whole team | Bigger potential gain | Higher cost and risk |
Check what you already pay for first. Many businesses hold AI features in existing plans without using them. Our guide to choosing the first process to automate helps you pick the target. If the question is build or buy, see build vs buy AI software for SMEs.
Step 5: Lay out the costs honestly
List every cost, including the ones people forget:
- Tool subscriptions. Use current prices from the vendor and say when you checked.
- Setup and build time, whether internal or an outside provider.
- Staff time for learning and testing.
- Ongoing review and maintenance.
- Anything you will stop paying for if this replaces it.
If an outside provider is involved, use their written quote. Put it in dollars, with GST shown (the ATO's GST guidance explains how GST works for businesses). Never hide costs to make the case look better. Partners notice, and one missing cost destroys trust in the rest of the document.
Step 6: State the benefits with your working shown
Convert your numbers into benefits and show the calculation:
Task: re-keying supplier invoices
Time now: 6 hours per week (sampled over 2 weeks)
Expected time after: 2 hours per week (estimate)
Hours freed: 4 hours per week
Cost of that time: $___ per hour (loaded)
Value per year: 4 x $___ x 48 = $___
Label estimates as estimates. Then name the benefits you cannot price: faster replies, fewer errors, less overtime, a better experience for the team. Be honest that time saved only has financial value if it is used for something useful, and say what that something is.
Never promise a specific saving that you cannot back up. A modest, credible claim is easier to approve than a dramatic one. If staff worry about their jobs, can AI replace my staff gives an honest answer you can share.
Step 7: Name the risks and the controls
This section builds trust. List the real risks and what you will do about each:
| Risk | Control |
|---|---|
| Sensitive data entered into a tool | Written AI use policy and an approved tool list |
| Wrong output reaches a customer | Human review before anything is sent |
| Staff do not use the tool | Pilot with willing users and training on real tasks |
| Costs grow | Review at set dates and a cap on seats |
| Tool or vendor changes terms | Check terms at each renewal and avoid lock-in where possible |
For the first row, see our guide to writing an AI use policy. Privacy and employment obligations vary, so check with an adviser where they apply. The OAIC guidance on commercial AI products and the Fair Work Ombudsman are the starting points for privacy and employment questions.
Step 8: Ask for a small first step
Frame the request as a trial with a defined end. For example: "Approve four weeks and a small budget to try this with three people on one process. We measure time and error rate against the baseline and review on [date]. If the target is missed, we stop."
A bounded request is easier to approve than an open-ended one. It also shows you are serious about checking results.
Step 9: Write it on one page
Use this order, with a short heading for each:
- The problem
- The proposal
- Options considered
- Costs
- Expected benefits
- Risks and controls
- The ask and the review date
Put the ask near the top as well, in one line, so the reader knows what decision you want. Then attach your data in an appendix if needed.
Step 10: Present it and listen
Walk the decision makers through the page and invite questions. Write down their objections, even the ones that seem unfair, and answer them in a follow-up. If they say no, ask what evidence would change their mind. That answer is your next step.
Copyable one-page template
BUSINESS CASE: [Title] Date: ____ Author: ____
THE ASK (one line): Approve $____ + GST and ____ weeks to trial ____.
1. PROBLEM
[Two sentences and one real example with numbers]
2. PROPOSAL
[What, for whom, replacing what]
3. OPTIONS
Do nothing / Use existing tools / Small trial / Larger rollout
4. COSTS
Tools ____ Setup ____ Staff time ____ Ongoing ____
5. BENEFITS
Priced: [hours x cost, working shown]
Unpriced: [list]
6. RISKS AND CONTROLS
[Risk - control, three to five rows]
7. MEASURES AND REVIEW
Baseline: ____ Target: ____ Review date: ____
Decision rule: keep / adjust / stop
What to do next
If you would like a second view on what is worth spending on, AI strategy from 21Labs works through your processes, tools and goals and gives you a clear plan and a priced first step. It starts with a free discovery. Send the form below and we reply within 24 hours.
Frequently asked questions
- How long should the business case be?
- One page for spend of a few thousand dollars, two to three for larger commitments. Decision makers want the problem, the ask, the cost, the benefit and the risk, in that order.
- What if I cannot put a dollar figure on the benefit?
- Use your own numbers with the working shown: hours per week, hourly cost, error rate, missed enquiries. Label estimates as estimates. Where a benefit cannot be priced, such as lower stress or faster replies, state it plainly instead of inventing a figure.
- Should I propose a trial or the full spend?
- Propose a small, time-boxed trial with a clear measure and a review date. It reduces the risk and makes the decision easier. Commit to the larger spend when the trial shows results.
- How do I handle a partner who is sceptical of AI?
- Ask what worries them and address it in the case. Common worries are data privacy, quality, cost and staff reaction. A short risk section with named controls usually does more than enthusiasm.
