AI Governance

AI System Card

A formal description of every AI system in the Metrixview platform — what each does, what data it accesses, and how human oversight is maintained.

Version 1.0 — Last updated 13 April 2026

1. Overview

Metrixview is a healthcare and governance performance management platform. It uses AI to provide bounded, framework-native governance intelligence — helping quality managers, executives, and administrators analyse organisational performance indicators through natural language.

All Metrixview AI systems operate at the governance layer: they process population-level indicator aggregates (e.g., readmission rates, compliance scores, quality metrics) for organisational decision support. They do not process individual patient records, individual clinical data, or personnel information.

Key principle: Every AI output in Metrixview is non-binding governance intelligence. No AI system makes automated decisions affecting individual rights. All outputs require human review and explicit action.

2. AI Systems Documented

Production

AI Data Chat

Purpose
Natural language governance analysis for quality managers and executives. Users ask questions about their indicator data and receive analytical responses.
Model
Google Gemini (paid tier) via the Mastra AI framework. Specific model version configured per environment.
Data accessed
Indicator tree structure, metric values, SPC (Statistical Process Control) summaries, annotations, and facility context (organisation name, facility name, sector, description).
Data NOT accessed
Individual patient records, clinical patient data, personnel records, financial systems, user credentials, or other organisations' data.
Output
Natural language responses — non-binding governance intelligence presented in a chat interface. No automated actions, alerts, or data modifications result from AI analysis.
Human oversight
All AI responses are presented as suggestions. Users must independently verify against source data and take any follow-up actions manually. No AI output triggers automated workflows.
Limitations
May produce plausible but incorrect analysis. Not a substitute for clinical judgement, regulatory compliance assessment, or professional advice. Users are advised to verify all outputs against source data.
Data retention
Conversation history is held in the user's browser session only. Not stored on Metrixview servers. Lost on page close or reload.
Production

AI Onboarding Wizard

Purpose
Assists new customers in defining their governance framework structure through a guided conversational wizard.
Model
Google Gemini (paid tier). Configuration: low temperature (0.3) for consistent, structured output.
Data accessed
User-provided input only: sector, organisation name, organisation size, role, goals, challenges, and optionally a website URL for additional context. No existing platform data is accessed.
Data NOT accessed
Existing indicator data, metric history, other users' data, patient data, or financial information.
Output
A proposed governance framework structure (groups and indicators) and a business analysis report. The framework is presented for user review — it is not imported automatically.
Human oversight
Mandatory user review and confirmation step. The generated framework is displayed for the user to inspect, modify, or reject before any import occurs. Import only proceeds after explicit user confirmation.
Data retention
Session messages are held in browser state during the wizard flow. The generated framework structure is stored in the database only after user confirmation.
Production

AI Observations

Purpose
Generates short narrative summaries (3–5 sentences) describing a facility's governance performance for inclusion in prospect sales reports (PDF).
Model
Google Gemini (paid tier) via the same agent framework as AI Data Chat.
Data accessed
Facility name and a summary of the top 8 indicators (name and current status) for the selected facility.
Data NOT accessed
Individual patient records, raw data points, other facilities' data, user credentials, or financial information.
Output
Static descriptive text included as one section in a PDF report. Non-binding narrative — no recommendations or automated actions.
Human oversight
Sales staff review the complete PDF before distribution to any prospect. The AI Observations section is one part of a larger report that requires human sign-off. If the AI fails to respond within 10 seconds, the section is omitted entirely.
Data retention
Generated text is cached in application memory for up to 24 hours per facility. Not persisted to the database.

3. Data Governance

3.1 Model Provider

Metrixview uses the Google Gemini API (paid tier) accessed through the Google AI Studio API endpoint (generativelanguage.googleapis.com).

No training on customer data. Under Google's paid-tier API Terms of Service, data submitted to the Gemini API is not used to train or improve Google's AI models. Data is processed under Google's privacy policy.

3.2 Conversation Log Retention

AI Data Chat and Onboarding Wizard conversation logs are not persisted to the Metrixview database. Conversation history exists in the user's browser session only and is lost on page close or reload. No server-side logging of AI conversation content occurs.

3.3 Data Isolation

Each AI API call contains only the data from the requesting user's current facility context. There is no cross-facility or cross-organisation data mixing in AI requests. Users can only query AI about data they already have access to through the platform's role-based access controls.

4. Regulatory Alignment

4.1 Privacy Act APP1 (Australia)

Metrixview's AI disclosure obligations under the Privacy Act APP1 amendment (effective 10 December 2026) are met through:

  • This AI System Card — public documentation of all AI systems, their scope, and safeguards.
  • The Metrixview Privacy Policy — Section 4 specifically addresses AI data handling.
  • In-app AI disclosure notice — first-use banner when opening AI Data Chat, linking to the Privacy Policy.

4.2 TGA Software as a Medical Device (SaMD)

Metrixview AI is not a Software as a Medical Device under TGA classification. Metrixview AI operates at the governance layer — processing population-level indicator aggregates for organisational quality management. It does not:

  • Provide individual clinical decision support.
  • Process individual patient records or clinical data.
  • Make or recommend treatment decisions.
  • Interface with clinical systems or electronic health records.

4.3 Automated Decision-Making (ADM)

Metrixview AI does not make automated decisions affecting individual rights. All AI outputs are governance intelligence for human review. Specifically:

  • No AI output triggers automated workflows, alerts, or notifications.
  • No AI output modifies data, statuses, or configurations without explicit user action.
  • No AI output is used to make employment, access, or service delivery decisions about individuals.

5. Contact

For AI governance enquiries, requests for additional information, or to report concerns about Metrixview's AI systems:

AI Governance Contact

Metrixview Pty Ltd

Email: [email protected]

Adelaide, South Australia

See also: Privacy Policy | Terms of Service