Building the case for an AI pilot in healthcare

In health, a strong business case for an AI pilot does more than ask for approval. It sets the benefits and the risks side by side, identifies accountability, and starts the work of governance before an AI pilot begins, while changing course is still cheap. Done well, it turns a promising idea into a decision leadership can make with confidence, and gives the pilot a safe path to scale.

This week I facilitated a 45-minute online session for the “Safe and Responsible Use of AI Working Group” at The Royal Victorian Eye and Ear Hospital. Members took one real AI use case through a seven-step business case that follows the Department of Health's Governance Models for AI Solutions in Victorian Public Health Services.

A business case template with assurance and built-in prompts lets anyone, even without being an AI expert, test whether an idea is worth doing, match governance effort to risk, name who is accountable and set the evidence needed to scale. The session also showed gaps no one had seen, such as unclear approval pathways.

Why build the business case first

A business case for an AI pilot forces the questions: who owns this, is it worth doing, how risky is it, what will we do about that risk, and what evidence would justify scaling? Answering them on paper, before a vendor demo or a pilot, is cheap, low risk and lean. Answering them after go-live is not.

To make that practical, I built a tool specific to health (I have built a similar AI business case tool for general sector use) that walks you through the seven steps below: Assurance to Pilot AI: a business case. It follows the Victorian Department of Health's Governance Models, Standards and Roles, Responsibilities and Escalation Pathways documents. It runs in your browser, so nothing leaves your approved environment, and it has no AI of its own. At each step it writes a prompt that you take to an approved AI, such as Microsoft Copilot.

The seven steps to build a business case for an AI pilot in health

1. Opportunity, accountability and decision makers. Name the problem, the person accountable and the committee that decides. Map your stakeholders here too: who are your allies, neutrals and sceptics (more on that below). This is where my research found the gap: accountability across the whole life of an AI. In the session, several members found they could not say which committee owns an AI decision. That is a useful thing to learn in a safe exercise.
2. Value triage (is this AI solution valuable enough to do the work of governance?). The Victorian Department of Health's guidance starts governance with this triage: is there a clearly defined problem and a measurable benefit that justifies the governance effort, and is the service ready to proceed? A "no" is a good result. It stops an idea before it consumes time.
3. Risk screen. Six risk questions set how much governance the idea needs. Higher risk, complexity or impact points to a fuller, multidisciplinary evaluation. Lower risk keeps the effort proportionate.
4. Benefits, risks and controls. A register of what you expect to gain, what could go wrong and what you will do about it.
5. Minimum governance. Every AI solution needs a baseline whatever its risk, built on approvals the service already runs: technology, safety, privacy, cyber security and clinical governance. The step shows what you already have and what is missing.
6. Scale through evidence. Plan a controlled pilot, evaluate it, harden it for production and extend it deliberately. Decide before you start what evidence would justify scaling. These principles come from MIT's Applied Agentic AI course.
7. The business case and committee report. Pull the steps together into one report for the decision makers, with the benefits and risk register and the stakeholder list.

What the room taught me

Know your stakeholders: are they a risk to the project, or people who can contribute? Naming allies, neutrals and sceptics is often overlooked in AI piloting and scaling. Members said the tool surfaced blockers they had not considered. Many of those blockers came from not knowing what the AI is for, not from informed opposition, so a sceptic is often someone who has not yet been shown the value. Scaling needs people to understand the value as well as the technology being ready, and the pilot, its testing, its governance and its ongoing promotion may each need different allies.

Honest gaps beat a blank page. When a regulatory or approval detail was unknown, the advice was to write "unknown", move on and come back. The gaps then become the to-do list.

Start from the most manual, repeated task. That question produced use cases quickly, from reconciling invoices and reviewing clinic schedules to preparing test scripts and reading handwritten patient questionnaires. Some were automation, not AI, and that is worth knowing early because the governance differs.

Governance work needs a route upward. A working group builds capability, but its output has to reach the organisation's risk register and the executive to count.

A guided path to a pilot you can defend

Describe the opportunity and the tool asks, one question at a time, who is accountable, whether the idea is worth the effort, how risky it is and what would justify scaling. Paste each reply back, check it, and the boxes fill. At the end it assembles the whole case into a report for your committee.

I deliberately did not build an AI agent, because a service will reasonably want to use only its own approved AI. Your answers stay on your own computer and nothing is sent to me. It is a planning aid, not an official Department tool and not a governance approval. It also draws on a related framework from Macquarie University and the Alfred that argues for tiered oversight and structured review.^[1]

Try it with your team

I can run this as a facilitated session, online or in person, for a working group, a digital leadership team or an executive. If you would like the tool for your own team, or a session built around your service, get in touch.

Thank you to the Safe and Responsible Use of AI Working Group and its chair for the invitation and the thoughtful discussion.

[1] Freeman S, et al. From principles to practice: an actionable framework for AI governance in healthcare organisations. BMJ Digital Health & AI 2026;2:e000018. Related published research; it is not cited in the Department of Health documents.
Department of Health: Artificial intelligence in Victorian public health services