Short answer
An AI business case is a short document asking a decision-maker to fund one AI project. It states the problem and its current cost, the options considered, one-off and running costs in NZD, the benefits you will measure, the risks and controls, and a stop rule. This template has ten sections that fit on two pages, with a filled-in example.
- Ask for one decision: a sum, a time period and a named workflow.
- Count review time and exceptions, or the benefits will be overstated.
- Hours saved are capacity, not cash, until you decide how to use them.
- Every business case needs a stop rule agreed before the pilot starts.
This template is for the person asking for money: an operations manager, practice manager or general manager who wants a director or board to approve one AI project. It assumes you have already chosen the workflow. If not, start with the AI readiness checklist. Copy the ten sections below into a document and fill in the brackets.
The AI business case template
1. Decision requested
One sentence: “Approve NZ$[amount] and [hours] of staff time for a [number]-week pilot of [what the AI does] in [workflow], reviewed on [date].”
2. Problem and current cost
- Workflow and owner: [name]
- Volume: [tasks per month], measured from [source] over [period]
- Time per task today: [minutes], timed on [number] real tasks
- Current cost: [hours per month] × NZ$[loaded hourly cost] = NZ$[amount] per month
- Other costs today: [errors, delays, backlog, lost work], with numbers where you have them
3. Proposed change
What the AI will do (read, extract, classify, draft, route), what people will still do, and the exact point where a person reviews the output before it takes effect.
4. Options considered
| Option | One-off cost | Monthly cost | Time to value | Main risk | Why chosen or rejected |
|---|---|---|---|---|---|
| Do nothing | NZ$0 | [current cost] | n/a | [risk] | [reason] |
| Use AI features in existing software | [ ] | [ ] | [ ] | [ ] | [ ] |
| Buy a specialist product | [ ] | [ ] | [ ] | [ ] | [ ] |
| Build or configure a custom system | [ ] | [ ] | [ ] | [ ] | [ ] |
5. Costs
| One-off | NZ$ | Monthly | NZ$ |
|---|---|---|---|
| Setup, build and integration | [ ] | Licences or subscriptions | [ ] |
| Testing and data preparation (staff time) | [ ] | Model and API usage | [ ] |
| Training | [ ] | Hosting and monitoring | [ ] |
| Contingency | [ ] | Support and maintenance | [ ] |
| Total one-off | [ ] | Total monthly | [ ] |
6. Benefits
Hours saved per month = (tasks × minutes today) minus (AI-handled tasks × minutes with AI including review) minus (remaining tasks × (minutes today + extra minutes per exception)), divided by 60. State what the saved time will be used for. List any quality or speed benefit only with the measure you will use.
7. Measures and baseline
| Measure | Baseline | Target | How measured | When |
|---|---|---|---|---|
| [e.g. minutes per task] | [ ] | [ ] | [ ] | [ ] |
| [e.g. share of outputs accepted with minor edits] | n/a | [ ] | [ ] | [ ] |
8. Risks and controls
The top three to five risks from your AI risk register, each with its control and owner. Always cover personal information, wrong outputs reaching customers, and supplier dependence.
9. Plan and stop rule
Phases with dates, the go or no-go review date, and the stop rule: “We stop if [measure] is below [threshold] at [date].”
10. Owners and approvals
Sponsor, workflow owner, privacy officer, finance. Each signs and dates.
How to fill it in
- Time the work before writing anything. Time 20 real tasks across a normal week. Estimates from memory are unreliable, and a business case built on a guess cannot be defended.
- Price every option, including doing nothing. Check what AI features you already pay for in Microsoft 365, Google Workspace or your industry software. MBIE’s quickstart for businesses using AI tools also asks you to understand future costs and the skills you will need (MBIE).
- Run the numbers. Use the AI ROI calculator for benefits and the AI cost calculator for model usage.
- Write a downside case. Re-run the numbers with AI handling a smaller share of tasks. If the case only works in the best case, propose a smaller pilot.
- Agree the stop rule with the approver. Before anyone starts building.
Filled-in example
Illustrative example: a 40-person Auckland freight forwarder handles 1,200 delivery exception emails a month. Each takes 9 minutes to read, check the shipment record and summarise for the customer service team. All figures are assumptions for illustration.
| Item | Value |
|---|---|
| Decision requested | NZ$25,000 and 60 staff hours for an 8-week pilot of AI-drafted shipment exception summaries, review on 15 March |
| Time today | 1,200 × 9 min = 180 hours a month |
| With AI | 75% drafted by AI and checked in 3 min; the other 25% take 9 + 2 min |
| Time after | (900 × 3) + (300 × 11) = 6,000 min = 100 hours |
| Hours saved | 80 a month at NZ$55 loaded cost = NZ$4,400 |
| Monthly running cost | NZ$700 (model usage, licences, monitoring and support) |
| Net monthly benefit | NZ$3,700 |
| One-off cost | NZ$25,000 (build and integration NZ$18,000, staff testing NZ$3,300, training NZ$1,700, contingency NZ$2,000) |
| Payback | 25,000 ÷ 3,700 = 6.8 months |
| Downside case | At 50% handled, saving is 40 hours, net NZ$1,500 a month, payback 16.7 months |
| Use of saved time | Clear the 3-day backlog; no change to roles |
| Stop rule | Stop if under 70% of drafts are sent with minor edits by week 6 |
The downside case changed the decision: the directors approved the pilot but asked for the 70% acceptance measure to be reported weekly.
Common mistakes
- Leaving out review time. If a person must check every AI output, that time belongs in the “with AI” minutes.
- Calling capacity cash. If saved hours mean fewer roles, say so plainly and plan the process: Employment New Zealand expects genuine business reasons, consultation and a fair process for changes that affect jobs (Employment NZ).
- Only one option. Approvers trust a case that shows the cheaper options it rejected and why.
- Forgetting running costs. Model usage, monitoring and someone to fix things when a supplier changes a model are monthly costs for as long as the system runs. Our guide to what custom AI projects cost lists the usual drivers.
- No stop rule. Without one, a weak pilot drifts into production because stopping feels like failure.
Next step
Once approved, turn sections 2, 3 and 7 into an AI project brief for whoever builds it, and set up the AI pilot scorecard so the review date has numbers ready. For the budget conversation itself, see creating an AI budget.
Sources
- Quickstart: Businesses using AI tools, Ministry of Business, Innovation and Employment, accessed 19 October 2026
- Responsible AI Guidance for Businesses, Ministry of Business, Innovation and Employment, July 2025
- The basics of workplace changes, Employment New Zealand, last modified 13 November 2025
Written by the Levy Street team in Wellington. Last updated 19 October 2026.