Free toolkit
AI business case toolkit
Last updated
A one-page business case, a vendor checklist and the security questions IT will ask, so you can take an AI pilot to your COO, CFO and CIO. Copy it, print it or forward it.
Who this toolkit is for
This toolkit is for the person who found the idea and now has to win approval for it. That is usually an application manager, IT lead, business analyst or continuous improvement manager.
Everything is on this page. There is no form and no download. Copy the parts you need into your own template, or print the page and forward it to the people who will decide.
The one-page AI pilot business case
A good business case for an AI pilot fits on one page. It names one process, measures it today, says what the AI would do and sets the result that means go or stop.
The example column below is illustrative only. It describes a fictional field service company with 12 branches. The numbers are round figures chosen to show the method, not results from any real organisation.
Business case template with an illustrative example
| Field | What to write | Illustrative example (fictional company) |
|---|---|---|
| Process name | One process, named the way staff describe it. Not a broad label such as AI for operations. | Turning completed job sheets into customer invoices. |
| Sites affected | Which sites run the process today, and the one site where the pilot starts. | All 12 branches. The pilot starts at one metro branch. |
| Volume per month | The count of transactions, documents or requests, taken from system reports. | About 4,000 job sheets a month across all branches. |
| Hours today | Time per item multiplied by volume. Measure a sample rather than estimate. | About 6 minutes per job sheet, or roughly 400 staff hours a month. |
| Error rate today | The share of items that need rework, a credit note or a follow-up call. | About 1 in 20 invoices is corrected after it is sent. |
| What AI would do | The specific step the AI performs, and where a person still approves. | Read each job sheet, draft the invoice lines and flag missing details. A coordinator approves every invoice before it is sent. |
| Who approves | The process owner, the budget holder and IT, by role or name. | Operations manager as process owner, CFO for budget, IT lead for security and integration. |
| Risks | The top three risks, each with the control that manages it. | Wrong parts charged: human approval of every invoice. Customer data exposure: data stays in Australia. Staff concern: coordinators help set the accuracy bar. |
| Cost to run | The pilot fee, then the monthly running cost per transaction, from written vendor quotes. | A fixed pilot fee and a monthly cost per job sheet, both from the vendor's written quote, compared with staff hours saved. |
| Measures of success | Two or three numbers, compared against the baseline above. | Time per job sheet down to about 2 minutes. Corrections below 1 in 50 invoices. |
| Go or stop criteria | The result that means scale, rescope or stop, and the date you will decide. | Go if drafts are at least 95 per cent correct on 500 job sheets over 6 weeks. Stop if below 85 per cent. Rescope in between. |
Tips for filling in the template
Measure before you write
A one-week sample of real work beats any estimate, and it becomes your baseline.
Start with one site
A single site keeps the pilot measurable and limits disruption if it needs to stop.
Keep a person in the loop
Approval of every output during the pilot answers most risk questions before they are asked.
Agree the go or stop rule early
Get the approvers to agree the numbers before the pilot starts, not after.
AI vendor evaluation checklist
| Area | Question to ask | What a good answer includes |
|---|---|---|
| Fit | Have you delivered this kind of process before, in an organisation of our size and structure? | Specific examples with similar volumes and several sites, not a general capability statement. |
| Fit | What would you not recommend we automate? | A clear view of where AI is a poor fit, including cases where a simple rule or report is enough. |
| Fit | Which of our systems would you need to work with? | Your actual systems, named after a short discovery, not a generic platform list. |
| Delivery approach | How do you scope the first phase? | One process, one site, a measured baseline and a fixed scope agreed in writing. |
| Delivery approach | What happens if the pilot misses the accuracy bar? | A go or stop decision point, with no obligation to continue into later phases. |
| Delivery approach | Who does the work on your side, and how much time do you need from our staff? | Named roles, a day-to-day lead and an estimate of hours needed from your team. |
| Security and data location | Where is our data stored and processed, including by any AI model? | Every location named in writing, in Australia, including backups and any model provider. |
| Security and data location | Is our data used to train any model? | No, confirmed in the contract and in the model provider's terms. |
| Security and data location | Which subcontractors or third parties can access our data? | A written list showing what each one does and where it is based. |
| Integration | How will the solution read from and write to our existing systems? | Supported APIs or approved connectors, with writing back to records added only after accuracy is proven. |
| Integration | What happens when one of our systems is upgraded or changed? | Who monitors each connection, how breakages are detected and who fixes them. |
| Ownership of code and IP | Who owns the code, prompts, configuration and models built for us? | You do, stated in the contract, with everything delivered into your environment or code repository. |
| Ownership of code and IP | Could our team run or change it without you? | Yes, with documentation, runbooks and handover training included in scope. |
| Ownership of code and IP | What happens to our data if the contract ends? | Return and deletion within a stated period, confirmed in writing. |
| Support after go-live | What support is included after go-live, and for how long? | Defined response times, a named contact and clear renewal or end terms. |
| Support after go-live | How do you monitor accuracy once it is live? | Regular checks against the agreed measures, with alerts when results drift. |
| References | Can we speak to a client whose project is in production today? | A reference willing to talk about what went wrong as well as what worked. |
| References | Can you show a real deliverable, such as a scope document or handover pack? | A redacted example that shows the quality of what you would receive. |
Security and data questions your IT team will ask
| Question IT will ask | What a good answer looks like |
|---|---|
| Does our data stay in Australia? | Yes. Data is stored and processed in Australian data centres, including by the AI model. Every location is named in writing, including backups. |
| How do users sign in? | Through your existing identity provider, with single sign-on and multi-factor authentication. No shared logins, and access ends when someone leaves. |
| What is logged, and who can see it? | Every AI action, input, output and human approval is logged with a user and a time. Logs are kept for an agreed period and your team can read them. |
| Is our data used to train AI models? | No. Training is switched off in the model provider's settings and excluded in the contract. Prompts and outputs are kept only as long as logging requires. |
| Who can access what? | The AI only sees data the user is already allowed to see. Permissions follow your existing roles, and service accounts have the least access they need. |
| How does this meet the Privacy Act? | Personal information is handled under the Australian Privacy Principles. Overseas disclosure is avoided or documented, a privacy impact assessment is done before go-live, and the breach response plan covers the Notifiable Data Breaches scheme. If the AI helps make decisions that significantly affect people, your privacy policy needs to say so under the Privacy Act transparency rules starting 10 December 2026 (see our automated decisions guide at /resources/privacy-act-automated-decisions/). |
| What happens if something goes wrong? | A named incident contact, a stated notification time and a way to switch the AI off while the manual process carries on. |
How to present it to leadership
Start with the problem in numbers
Open with hours, errors and backlog today. Do not open with AI.
Show one small pilot
One process, one site, a fixed scope and a decision date.
Name the risks and controls
Cover data location, human approval and logging before anyone asks.
Put cost and measures side by side
What the pilot costs, what it must prove and how you will know.
Ask for one decision
Approval for the first phase only, with a go or stop review at the end.
What each executive needs to hear
| Executive | What they need to hear | Bring this |
|---|---|---|
| COO | The process gets faster and more consistent across sites without disrupting operations. Staff still approve the output. | The baseline hours and error rate, and the site where the pilot starts. |
| CFO | Cost is capped by phase, the measure is agreed up front and there is a clear point to stop. | The pilot cost, the running cost per transaction and the go or stop criteria. |
| CIO | Data stays in Australia, access follows existing permissions and the organisation owns the code. | The vendor's answers to the security questions above and the integration approach. |
How Solve8 can help a champion
If you are building the case for an AI pilot, we can review it with you in a 30-minute call. Bring your draft, or just the process you have in mind.
Afterwards you receive a written one-page summary covering the process, the likely data and integration questions, the main risks and a sensible first phase. It is written to be shared with your COO, CFO and CIO, and it does not commit you to anything.
Questions
What should a one-page AI business case include?
A one-page AI business case names one process, its monthly volume, the hours and error rate today, what the AI would do and who approves it. It adds the top risks, the cost to run, two or three measures of success and the result that means go or stop.
What questions should you ask an AI vendor?
Ask where your data is stored and processed, whether it trains any model, who owns the code and what support continues after go-live. Then ask for a client reference in production and what the vendor would not automate. Vague answers to these are a warning sign.
How do you measure ROI on an AI pilot?
Measure the process before the pilot starts, then compare the same numbers afterwards. Typical measures are staff hours per transaction, error or rework rate and turnaround time. Weigh the gain against the pilot fee and the monthly running cost, including model usage at your real volumes.
How do you get executive approval for an AI project?
Lead with the business problem in numbers, then ask for a small, capped first phase with a clear go or stop point. Each executive needs a different answer: operational impact for the COO, capped cost for the CFO, and security and ownership for the CIO.
What security questions should IT ask before approving an AI tool?
IT should confirm where data is stored and processed, how users sign in, what is logged, whether data trains any model and how access permissions are enforced. They should also check Privacy Act obligations, including overseas disclosure and breach notification.
Get your business case reviewed
A 30-minute review gives you a written one-page summary to share with your leadership team.
Fixed-fee phases, agreed before we start. Stop after any phase.