AI in Healthcare in New Zealand: A Living Whitepaper

New Zealand healthcare is moving from AI pilots to supervised deployment, led by documentation tools. The next bottleneck is safe integration: shared data, procurement, equity, cyber resilience and evidence of patient benefit.

Executive Summary

AI adoption in Aotearoa New Zealand healthcare continues to expand, but the pattern remains selective, supervised and infrastructure-dependent.

The period since the previous update has not produced another national clinical AI go-live comparable with the emergency-department scribe rollout. Instead, the most important developments have been:

• Health New Zealand’s proposed national radiology platform has entered a procurement pathway that explicitly includes an imaging-AI orchestration layer. The opportunity remains subject to approval and is not evidence of deployment. (gets.govt.nz)
• Health NZ has moved the Shared Digital Health Record into early-adopter onboarding, with primary-care data sharing still expected to begin from mid-2027. (healthnz.govt.nz)
• Four major primary health organisations have called for a nationally aligned digital-health strategy, stronger interoperability and greater investment in AI and digital tools. This is sector advocacy, not government policy. (hinz.org.nz)
• New University of Auckland research has highlighted the limited readiness and uncertain equity of AI tools for pain assessment in older adults. Only 5.5% of tools reviewed were considered ready for clinical integration, and none of the reviewed studies were conducted in New Zealand. (auckland.ac.nz)
• AI governance is becoming more distributed across professional regulators, including the Medical Council, Psychologists Board and Dietitians Board. Their common position is that AI may support practice, but accountability remains with the practitioner. (mcnz.org.nz)

The adoption hierarchy remains broadly unchanged:

1. Documentation and workflow assistance: most mature, with national public-sector emergency-department use and growing adoption in primary care and aged care.
2. Patient engagement: early deployments in health coaching and digital navigation.
3. Imaging and screening: moving through procurement, equipment upgrades and local validation.
4. Clinical decision support: mainly research and controlled evaluation.
5. Autonomous diagnosis or treatment: not supported by current Health NZ operating guidance.

The strategic question is therefore shifting from whether AI can be introduced to whether New Zealand can provide the data, governance, procurement and workforce conditions needed to operate it safely at scale.

What Has Changed Since the Last Update

Radiology has become the clearest infrastructure-scale AI pathway

On 28 August 2026, Health NZ’s Future Procurement Opportunity notice described a proposed Radiology Modernisation Programme valued at $50 million to $100 million. The programme would procure an enterprise imaging integration layer, an integrated reporting environment, a national image archive and an Imaging AI Orchestration system capable of connecting radiology workflows to AI applications. (gets.govt.nz)

This is strategically significant because it treats AI as part of a national clinical platform rather than as a series of disconnected departmental tools. It could allow different imaging-AI applications to be assessed and introduced over time without rebuilding the underlying data and workflow architecture for every use case.

The status must nevertheless be stated clearly:

• The notice is a future procurement opportunity.
• The project was listed as awaiting approval.
• Health NZ explicitly said the notice did not commit it to procure the stated goods or services.
• A separate national radiology-information-system procurement is expected later.

The development is therefore evidence of procurement intent and architectural planning, not evidence that a national imaging-AI service is operating.

Shared data infrastructure has moved closer to implementation

Health NZ has begun onboarding early-adopter practices in Auckland, New Plymouth, Masterton, Taupō and Ashburton to the Shared Digital Health Record. Participating practices are testing agreements, security due diligence, patient communications and onboarding processes. Primary-care data sharing is expected to begin from mid-2027. (healthnz.govt.nz)

This matters for AI because more advanced applications depend on longitudinal information that is currently distributed across general practices, hospitals, pharmacies, urgent-care services and national datasets. The Shared Digital Health Record is not itself an AI system, but it is part of the data layer required for:

• Cross-setting decision support.
• Safer clinical summarisation.
• Population-risk analysis.
• Care coordination.
• More complete evaluation of AI performance across patient groups.

Health NZ’s onboarding model also shows that interoperability is being treated as a security and governance problem, not simply a technical connection. Practices are being asked to complete cyber-security and due-diligence checks before participation. (healthnz.govt.nz)

Primary-care organisations are pushing for a stronger national direction

On 24 August 2026, the Network4 Alliance—representing Pegasus Health, Pinnacle Group, ProCare and Tū Ora Compass Health—called for a nationally aligned digital-health strategy with primary care at its centre.

Its election manifesto calls for:

• Greater investment in AI and digital tools.
• A single, secure health record.
• National interoperability standards.
• More effective scaling of New Zealand-built platforms.
• Technology investment tied to patient and clinician outcomes.

The alliance represents more than 350 general practices and over 1.8 million enrolled people, giving the position substantial sector weight. However, it remains an advocacy document rather than a government commitment or funded implementation plan. (hinz.org.nz)

The timing is important. It indicates that primary care increasingly sees AI adoption as constrained by national architecture, funding and implementation support—not simply by the availability of commercial tools.

New evidence has sharpened the equity question

A University of Auckland study published on 28 August 2026 reviewed 96 studies of AI for pain assessment and management in older adults. Nearly 76% of the tools remained at proof-of-concept stage, only 5.5% were judged ready for clinical integration, and none of the reviewed studies had been conducted in New Zealand. (auckland.ac.nz)

The study also found that two-thirds of the research involved mixed-age populations rather than tools designed specifically for older adults. The authors warned that systems developed overseas may not perform equally well for older people, Māori, Pacific peoples or other populations whose language, culture and patterns of communication may be under-represented in training data.

This is not a study of an operating New Zealand AI system. Its importance is evaluative: it demonstrates how quickly claims about technical possibility can outrun evidence of clinical readiness and local suitability.

Industry capability is becoming more nationally coordinated

Industry reporting on 31 August 2026 said Health NZ had appointed Gary Baird as National Director, Digital Services and Technology, in a newly created national role. The role is intended to bring together local expertise and national leadership across digital services. (pulseit.news)

The appointment is not an AI deployment. Its significance is organisational. AI projects are more likely to scale when responsibility for platforms, security, architecture, procurement and clinical digital services is coordinated rather than left to separate regional initiatives.

Similarly, a national partnership announced by Health Accelerator and Heidi on 20 August 2026 is evidence of a broader commercial push into general practice, but not proof that all participating practices have deployed the technology. (healthaccelerator.co.nz)

Current State of AI Adoption

Documentation remains the leading operating use case

The clearest example of AI operating at national scale remains Heidi’s deployment across public emergency departments.

Health NZ reported that access had been completed across all emergency departments, reaching approximately 1,250 doctors and frontline staff. The Government cited pilot and early rollout feedback indicating reduced documentation time, the potential to see an additional patient per shift and positive effects on clinician experience. These results are based on pilot findings and staff feedback; they are not yet equivalent to an independent national evaluation of patient outcomes. (beehive.govt.nz)

The operating model remains bounded:

• The system drafts documentation from clinical conversations.
• Clinicians review and approve the output.
• The clinician remains responsible for the final record.
• The tool is not authorised to make autonomous diagnostic or treatment decisions.
• Consent, privacy and secure handling of recordings remain central requirements.

The national rollout therefore represents a substantial administrative deployment, not autonomous clinical care.

Primary care adoption is expanding faster than standardisation

Primary care is likely the largest market for AI scribes and related workflow tools, but national adoption data remains incomplete.

Medtech AI is marketed as a New Zealand healthcare documentation platform integrated with Medtech Evolution. Its patient information describes real-time transcription, structured notes, referral letters and patient summaries, with clinician approval required before anything is written to the record. Medtech also states that the service is used by general practices, specialists, allied-health providers, mental-health clinicians and urgent-care services. These are provider-reported deployment claims rather than independently audited adoption statistics. (medtechglobal.com)

The broader implementation picture is uneven. A New Zealand primary-care survey cited in the previous edition found substantial AI-scribe experience among respondents, alongside continuing uncertainty about consent, vendor terms, legal compliance and local governance. The latest sector activity does not yet show that these gaps have been resolved.

The practical distinction is between:

• Tool availability: a vendor or practice can purchase access.
• Operational use: clinicians regularly use the tool in live consultations.
• Safe adoption: the practice has consent, training, review, incident management, privacy and security processes.
• Measured value: the organisation can demonstrate improved workload, care quality, access or outcomes.

Much of the market is currently between the second and third categories.

Aged care is a meaningful extension of the scribe model

Metlifecare’s rollout of HEIDI beyond its initial Parkside Village trial shows that ambient documentation is spreading into aged residential care.

The initial trial involved 12 registered nurses over four months. Metlifecare subsequently announced deployment across three additional care homes, reporting less administrative time and more opportunity for resident care and whānau conversations. Staff review and approval of generated notes remains required. (metlifecare.co.nz)

The case is important because aged care has different documentation, consent and communication requirements from emergency medicine or general practice. Residents may have cognitive impairment, family members may be involved in care, and conversations can include sensitive information beyond a standard consultation. These factors make implementation design as important as transcription accuracy.

Patient-facing coaching is still early

Tāmaki Health’s AI Health Coach, developed with Groov, is being rolled out in selected clinics to support behaviour change and self-management in areas including diabetes, gout, weight management and anxiety. The tool is designed to operate between appointments and escalate users to human health coaches or clinical teams when needed. (healthcareitnews.com)

The model is intentionally less clinically autonomous than a diagnostic chatbot. It provides coaching and action planning rather than diagnosis or treatment selection.

Reported results from earlier conversations include an average helpfulness rating of 8.5 out of 10 and improved self-reported confidence. These are provider- and vendor-reported indicators. There is not yet sufficient independent evidence that the service improves sustained health outcomes, reduces inequities or lowers demand without creating new risks.

Imaging is moving from isolated capability to national platform planning

Health NZ’s proposed radiology programme is the main national pathway for imaging AI. It would create the technical environment for image sharing, reporting, archiving and connection to AI applications. (gets.govt.nz)

There are also local examples of AI embedded in imaging equipment. Health NZ has reported that a new SPECT-CT scanner in Southland uses AI-assisted image reconstruction to produce clearer images and support faster clinical decision-making. This is an operating equipment capability, but it should not be confused with an autonomous diagnostic system: the technology assists image production, while clinical interpretation remains with health professionals. (healthnz.govt.nz)

National breast-screening AI remains a validation and procurement pathway rather than an established autonomous diagnostic service. The proposed model continues to place clinicians in responsibility for diagnosis, follow-up and treatment decisions.

Clinical decision support remains research-led

New Zealand’s most ambitious clinical-AI activity remains concentrated in research, trials and proof-of-concept work.

The REVOLUTION trial described in the previous edition is testing machine-learning-guided oxygen therapy for critically ill patients. Its significance is that it evaluates treatment outcomes rather than simply measuring whether an algorithm predicts deterioration accurately.

This research-led approach is appropriate for high-risk use cases. A model that performs well retrospectively may still fail when clinicians respond to it, when patient populations change, or when the recommendation alters the clinical workflow. Evidence of benefit must therefore include safety, implementation, equity and patient outcomes.

Governance, Policy and Regulation

Health NZ is formalising pre-implementation evaluation

Health NZ’s national digital-technologies guidance now identifies a framework for a consistent approach to evaluating AI tools before implementation. The framework is associated with the University of Auckland’s TRANSFORM programme and the National Artificial Intelligence and Algorithm Expert Advisory Group. (healthnz.govt.nz)

The direction of travel is toward repeatable assessment of:

• Clinical safety and effectiveness.
• Privacy and data security.
• Equity, bias and cultural appropriateness.
• Workflow fit and interoperability.
• Evidence of benefit.
• Procurement and implementation readiness.
• Monitoring after deployment.

This is a shift from individual projects relying primarily on local champions or vendor assurances. It does not eliminate local clinical responsibility, but it should reduce variation in how proposed systems are assessed.

Professional accountability is becoming explicit

The Medical Council’s March 2026 guidance states that doctors must be satisfied that AI is safe and suitable, remain responsible for clinical decisions, obtain consent in relevant situations, consider bias and ensure adequate privacy and security safeguards. (mcnz.org.nz)

The New Zealand Psychologists Board updated its AI guidance in July 2026. The revised guidance reinforces practitioner accountability, the difference between a system having knowledge and having understanding, the need to review outputs for bias and error, and the importance of informed choice for service users. (psychologistsboard.org.nz)

The Dietitians Board has also consulted on draft AI guidance. The draft applies across clinical care, documentation, education, research and service development. It requires critical review of outputs, privacy and security safeguards, appropriate disclosure and consideration of Māori data sovereignty, cultural safety and language. (dietitiansboard.org.nz)

Across the professions, the emerging standard is consistent:

AI may assist professional practice, but responsibility cannot be delegated to the system.

Privacy and security risks now affect procurement confidence

The Office of the Privacy Commissioner’s findings on the Manage My Health breach are not specific to AI, but they are directly relevant to AI adoption. The inquiry found that 99,416 patients were affected and concluded that both Manage My Health and Health NZ had failed to maintain reasonable security safeguards. It also warned against relying solely on vendor assurances and recommended a centralised programme to verify key health-sector suppliers. (privacy.org.nz)

For AI procurement, the implications include stronger scrutiny of:

• Multifactor authentication.
• Identity and access management.
• Data retention and deletion.
• Subprocessors and cloud hosting.
• Contractual accountability.
• Security testing and incident response.
• Audit logs and post-deployment monitoring.

The Health Information Privacy Code remains a core legal framework for health-data use. Its 2026 amendments, including changes associated with Information Privacy Principle 3A, reinforce the need for health agencies and suppliers to understand their obligations when collecting and using personal information. (privacy.org.nz)

The digital foundation is not yet complete

Health NZ’s Shared Digital Health Record is still in onboarding and readiness testing. Immunisation and medication data are expected to become available in late 2026, while primary-care data sharing is expected from mid-2027. (healthnz.govt.nz)

This creates a practical limit on advanced AI. Tools that depend on complete longitudinal records cannot perform reliably when information is missing, inconsistently coded or unavailable across care settings.

The foundation also has to support patient choice. Health NZ says people will have options over how much information is shared, while providers and systems must meet security, authentication, monitoring and access-control requirements. (healthnz.govt.nz)

Case Studies

Case Study 1: Heidi in emergency departments

Use case: Ambient transcription and clinical documentation.

Status: Operating nationally across public emergency departments.

Reported benefits: Reduced documentation time, greater clinician capacity and improved staff experience.

Evidence status: Early results are primarily government, provider and staff-reported. Independent evaluation of patient outcomes, documentation errors and long-term clinician behaviour remains limited.

Implementation lesson: National access does not remove the need for human review, consent, incident reporting, adversarial testing and clear limits on use. (beehive.govt.nz)

Case Study 2: Health NZ radiology modernisation

Use case: National imaging integration, reporting, archiving and orchestration of imaging-AI applications.

Status: Future procurement opportunity; project awaiting approval.

Potential value: More consistent access to images, less duplication, improved reporting workflows and a platform through which multiple AI applications could be evaluated.

Evidence status: No operating national AI deployment has been established through this procurement notice.

Implementation lesson: The health system is beginning to treat AI capability as dependent on national architecture, data standards and workflow integration rather than as a standalone software purchase. (gets.govt.nz)

Case Study 3: Tāmaki Health and Groov

Use case: Patient-facing health coaching between appointments.

Status: Early rollout in selected clinics.

Reported benefits: Personalised action plans, continuous digital access and escalation to human health coaches.

Evidence status: Early engagement and helpfulness measures are self-reported and vendor-generated.

Implementation lesson: Patient-facing AI requires stronger safety-netting and escalation than administrative tools, particularly when supporting mental health or chronic disease management. (healthcareitnews.com)

Case Study 4: AI-assisted imaging in Southland

Use case: AI-assisted image reconstruction within a SPECT-CT scanner.

Status: Operating locally.

Reported benefits: Clearer images, broader imaging capability and potentially faster clinical decisions.

Evidence status: This is an equipment-level AI capability, not autonomous diagnosis.

Implementation lesson: AI can enter clinical services through imaging hardware and workflow tools even when national diagnostic-AI procurement remains at an earlier stage. (healthnz.govt.nz)

Case Study 5: AI in older-adult pain care

Use case: Pain assessment and management using facial, voice, image, video and other data.

Status: Global research landscape reviewed by University of Auckland researchers; no New Zealand implementation identified in the review.

Key finding: Most tools remain at proof-of-concept stage, with very few judged ready for clinical integration.

Implementation lesson: Older adults, Māori and Pacific populations should be involved in design and validation from the beginning, rather than treated as a later adaptation problem. (auckland.ac.nz)

Trends

1. Infrastructure is becoming the central adoption issue

The next phase of AI adoption will depend less on access to models and more on:

• Interoperable records.
• National imaging infrastructure.
• Secure APIs.
• Data standards.
• Identity and access controls.
• Reliable clinical workflows.

The radiology programme and Shared Digital Health Record illustrate this shift.

2. Documentation is scaling because it is comparatively bounded

Scribes have a clear value proposition and can operate with a clinician-editor model. They reduce administrative work without formally transferring diagnosis or treatment responsibility to an algorithm.

This makes them easier to approve than tools that influence triage, diagnosis, treatment selection or patient self-management.

3. Patient-facing AI is the next governance test

Health coaching, symptom checking and mental-health triage all place AI closer to the patient. They create higher requirements for:

• Clear scope and safety-netting.
• Reliable escalation.
• Accessible alternatives.
• Cultural and linguistic fit.
• Monitoring for inappropriate reassurance.
• Evaluation of behaviour and health outcomes.

The evidence base is currently thinner than the level of interest.

4. Equity is moving from principle to evaluation criterion

The older-adult pain review, diabetic-retinal-screening experience and professional guidance all point to the same conclusion: imported models cannot be assumed to perform equitably in Aotearoa.

Evaluation must address:

• Māori and Pacific performance.
• Older people and disabled people.
• Rural and low-connectivity settings.
• Te reo Māori and Pacific languages.
• Different communication styles.
• Digital exclusion.
• Māori data sovereignty.

5. Procurement announcements must not be mistaken for deployments

The sector now contains several different evidence categories:

• Announced funding.
• Future procurement opportunities.
• Proofs of concept.
• Local pilots.
• Operating deployments.
• Scaled services with measured outcomes.

These categories should be reported separately. The proposed radiology-AI orchestration layer is a good example: it is strategically important, but it is not yet a live national service. (gets.govt.nz)

6. Professional regulators are filling policy gaps

New Zealand does not yet have a single health-specific AI statute governing all clinical applications. Professional regulators are therefore establishing practical standards through guidance on competence, consent, privacy, bias, oversight and accountability.

This is useful, but it also creates the risk of inconsistent expectations across professions unless Health NZ and the Ministry provide stronger cross-sector coordination.

Outlook

Over the next 12 to 18 months, the most plausible developments are:

• Further expansion of AI scribes in primary care, mental health and aged care.
• Continued onboarding and testing for the Shared Digital Health Record.
• Progress toward national radiology-platform procurement.
• Local imaging-AI use through new scanners and radiology systems.
• Evaluation of patient-facing coaching and symptom-checker models.
• Development of AI-enabled mental-health triage and referral.
• Wider use of Health NZ’s national pre-implementation framework.
• More professional guidance on AI competence and informed consent.
• Greater scrutiny of vendor security, data residency and contract controls.

The least likely near-term outcome is broad autonomous diagnosis or treatment. Current Health NZ guidance, professional standards and the adoption pattern all favour clinician-supervised augmentation rather than delegation of clinical responsibility. (healthnz.govt.nz)

The critical test will be whether the health system can turn infrastructure investment into measurable public value. Faster notes and better images may be useful, but stakeholders will increasingly need evidence about patient access, clinical quality, equity, safety, workforce wellbeing and total cost.

Overall Assessment

As of 1 September 2026, AI adoption in New Zealand healthcare is progressing, but the latest movement is more institutional than spectacular.

The sector has:

• A nationally operating AI documentation service in emergency departments.
• Expanding use of scribes in primary care and aged care.
• Early patient-facing coaching.
• Local AI-assisted imaging capabilities.
• A national radiology platform in procurement planning.
• Research into AI-guided clinical treatment.
• National and professional AI-governance frameworks.
• Shared-data infrastructure moving through early-adopter onboarding.

The main constraint is no longer a lack of interest or commercial supply. It is the health system’s ability to integrate AI into secure, interoperable, equitable and clinically accountable services.

The strongest conclusion remains:

AI is assisting healthcare work, but clinicians and health organisations remain responsible for deciding where it is safe, useful and appropriate.

New Zealand’s progress should therefore be measured not by the number of AI announcements, but by the number of systems operating reliably in real clinical settings, supported by independent evidence and trusted by patients, whānau and health professionals.