Queensland Health Artificial Intelligence Framework S3: Stage 3, business case (Gate 2): fully develop the preferred option with reconfirmed governance and team, a risk management plan for all risks to date, financial controls and funding, stakeholder analysis, regulatory and legal approval pathways such as TGA, model validation criteria and integration capacity; for in-house or partnership builds assess skills, authorise platforms and data sharing, define and split datasets, design architecture, co-design with end users, train with interpretability and bias mitigation, validate on unseen data, assess impacts and report with a model card; for procurement establish commercial expertise, approve the procurement strategy, evaluate vendors on healthcare AI experience, privacy and security, interpretability and bias mitigation, support and due diligence, and set evaluation criteria; then evaluate performance against predefined thresholds covering performance, interpretability, fairness, safety, control, security, privacy, robustness and reproducibility and across diverse consumers, establish total cost of ownership, document datasets and parameters, draft deployment, validation, retraining and continuity specifications, load testing and integration design, update the risk assessment and consult change management plans

What does S3 require?

Stage 3 fully develops the preferred option and tests whether it meets business objectives and expectations before proceeding. Irrespective of pathway the initiative owner reconfirms or establishes governance and the project plan, sets up the team with the skills and management expertise needed, completes a risk management plan for every risk identified so far with measures to minimise, mitigate or manage them, defines financial controls, funding and resources, analyses stakeholder consultation and its effect on the solution, researches and defines the regulatory or legal approvals required such as TGA software as a medical device approval, defines criteria or targets for successful model validation, and assesses capacity and processes to integrate the solution including workflow mapping, consultation and change management strategies. For in-house development or partnerships the team self-assesses its technical skills and support, determines data and model platform authorisation for external users and sharing mechanisms, defines the development dataset including training and testing splits and exploratory analysis, designs the model architecture choosing techniques suited to the problem, data and performance needs, co-designs functionality and interface with end users, trains the model while assessing interpretability and mitigating bias, validates with techniques such as cross-validation on data not used in development, assesses the impact on patients, workforce, environment and society and other identified risks, and reports pilot, trial or prototype results in a model card. For procurement the team establishes commercial expertise on supplier capability, track record and fit, gains approval of the procurement strategy, evaluates vendors on healthcare AI experience, data privacy and security practices, interpretability and bias mitigation, performance and post-implementation support and financial, business and legal due diligence, and sets vendor evaluation criteria. For both pathways the team evaluates the solution against predefined thresholds for model performance, interpretability and explainability, fairness, safety, control, security, privacy, robustness and reproducibility (revisions escalated to governance), evaluates performance for health consumers from diverse backgrounds to ensure equity, establishes total cost of ownership, documents the useful datasets, best model and parameters and any architecture or pipeline changes, drafts deployment specifications covering batch or real-time inference, transition validation, retraining frequency and business continuity, determines performance tuning and load testing, obtains approval of the integration design (container, API or standalone front end), updates the risk assessment, and develops and consults change management plans including training, reskilling and any industrial relations steps. Governance with or able to access subject matter expertise then approves progression to Stage 4, straight to deployment, approval pending funding and priority, or not.