From Connected Infrastructure to Trusted Intelligence: Making Government 3.0 Work in Practice

1. Executive Summary

The discussion explored how public sector organisations can move from fragmented digital services toward Government 3.0: a connected, interoperable and journey-led model of government. Participants framed Government 3.0 as the necessary foundation before more advanced AI-enabled or “Government 4.0” capabilities can operate safely and at scale. While the conversation touched on identity layers, data exchange, AI use cases and digital infrastructure, the strongest insight was that technology alone is not the constraint. The harder challenge is organisational: culture, incentives, trust, governance, capability and shared purpose.

Participants highlighted that many agencies are already experimenting with AI, automation and data-driven workflows. However, the transition from pilot to production remains difficult. One participant noted the familiar problem that around 70% of pilots do not scale, not because the technology fails, but because organisations cannot embed the innovation into operating models, workflows and cross-agency structures.

The brief identifies several practical lessons: start with problem definition rather than technology mandates; prioritise “unsexy” back-end automation where value and risk are clearer; build AI literacy across executives and delivery teams; treat responsible AI as an operational capability; and create shared value propositions that give agencies a reason to collaborate.

The central conclusion is clear: Government 3.0 is not just a digital architecture. It is a trust architecture.

2. Key Themes and Insights

Government 3.0 is the Digital Road Network, Not the Destination

Participants described Government 3.0 as the enabling layer for a more connected public sector. Government 0.0 was characterised as paper-based administration. Government 1.0 digitised forms and basic web interactions. Government 2.0 introduced portals and app-based services, but often left services fragmented behind the scenes. Government 3.0, by contrast, is about connected government: shared identity, interoperable data, journey-led services and reduced friction for citizens and service providers.

One participant compared Government 3.0 to the physical equivalent of digital roads. Roads do not themselves create innovation, but they allow people, services and markets to connect. In the same way, interoperable infrastructure creates the conditions for scalable digital government. Without it, promising innovations remain isolated within departments, programs or pilots.

This framing is important because it positions Government 3.0 as a necessary precondition rather than a final ambition. Advanced AI, agentic services and hyper-personalised citizen experiences require trusted data flows, clear permissions and consistent operating models. Without those foundations, Government 4.0 risks becoming a collection of disconnected experiments layered onto legacy fragmentation.

Connection is Necessary, But Not Sufficient

A central theme was that connected systems do not automatically produce collaborative outcomes. Participants stressed that interoperability is not only a technical issue; it also depends on behaviours, incentives, governance and culture. As one participant put it, connection is critical, but it is not sufficient.

This distinction matters. Governments often approach digital transformation as a technology integration challenge: connect the systems, expose the APIs, build the platforms. Yet participants repeatedly returned to the human and organisational layers: how people communicate, how roles are designed, how incentives are set, how risk is shared and how decisions are made.

The discussion also highlighted the difference between visibility and scale. Knowing that another team or agency has built something useful is not the same as having the mechanisms to adopt, fund, assure and operationalise it. This is why pilots frequently fail to scale. The technology may work, but the organisation is not ready to absorb it.

Government 3.0 therefore requires systems integration in the broadest sense: technology, data, people, governance and operating models moving in the same direction.

The Real Barrier to Scale is Operating Model Fit

The group returned several times to the challenge of moving from innovation to transformation. Participants noted that government and university settings often produce promising pilots, proofs of concept and local automations. However, these rarely become system-wide capabilities.

The issue is not always access to tools. In many cases, the tools are available, the technology is capable and the local use case has been proven. What is missing is the organisational machinery required to scale: shared funding, workforce capacity, executive alignment, data readiness, assurance processes and a common value proposition.

One participant described this as the challenge of turning innovation into transformation. Another noted that 70% of pilots do not scale, largely because organisations cannot take them across the enterprise.

This is particularly relevant for AI. Many agencies are experimenting with AI scribes, workflow automation, decision support, data exchange and productivity tools. But without clear prioritisation, data quality, AI literacy and responsible governance, pilots risk becoming either performative innovation or unmanaged risk.

AI Mandates Can Distort Problem-Solving

A strong critique emerged around the tendency for boards or executives to mandate AI adoption before defining the problem. Participants described pressure to be “seen to be innovative”, with AI sometimes treated as the objective rather than a possible method.

This creates several risks. First, it can divert scarce workforce capacity away from more valuable operational improvements. Second, it can push agencies into compressed timelines that weaken due diligence, assurance and evaluation. Third, it can generate solutions looking for problems, rather than targeted interventions tied to measurable outcomes.

The health sector examples were especially instructive. AI scribes were identified as a potentially valuable use case, particularly because clinicians are already using tools in private practice and want relief from documentation burden. However, participants also warned that rushed implementation could bypass proper evaluation, ethics, privacy and workflow design.

The better approach is to start with service outcomes: reducing cost per transaction, improving quality, shortening wait times, reducing clinician frustration, improving continuity of care or lowering administrative burden. AI should then be tested against those goals, not adopted for its own sake.

3. Innovative Ideas and Case Studies

Back-End AI: The Unsexy Work That May Deliver the Most Value

One of the most useful distinctions in the discussion was between front-end AI and back-end AI. Participants noted that public attention often focuses on citizen-facing tools, chatbots and hyper-personalised services. Yet many of the highest-value use cases may sit behind the scenes.

Examples included AI-driven workflows for sorting CVs, answering enterprise agreement questions, offboarding staff from multiple systems and checking Medicare eligibility before appointments. These are not headline-grabbing innovations, but they reduce administrative friction and improve operational accuracy.

Another participant described a transport-related data-sharing use case where major incidents on motorways need to be reflected quickly in mapping services such as Google Maps, Apple Maps and Waze. To meet tight service expectations, the state must detect and classify incidents rapidly, drawing on camera feeds, IoT data and integrated systems. AI can support this kind of back-end detection and routing without directly replacing the visible human service layer.

The broader lesson is that AI may deliver the greatest early public sector value when it reduces waste, improves responsiveness and supports existing service obligations — especially in high-volume, low-visibility processes.

Health Data as an Early Warning System

Health-related examples illustrated the potential of AI, data exchange and predictive analytics when reliable data is available. One participant described using symptom checker data from around 2 million people per year to support surveillance. Because people often seek symptom advice before presenting to doctors or emergency departments, this data can provide signals up to two weeks ahead of Medicare or hospital presentation data.

Another example involved infection surveillance at Peter Mac, where relatively simple natural language processing was used to detect histopathology and CT scan reports that suggested infection. These results were surfaced in a dashboard so infection teams could monitor burden and trends.

These cases show that sophisticated public value does not always require highly complex models. Sometimes the value comes from connecting existing data, applying targeted analytics and getting the right information to the right teams early enough to act.

Predictive Emergency Department Wait Times

Participants also discussed AI modelling of emergency department wait times. At Royal Melbourne Hospital, data on how long people have been waiting in ED is used to provide anticipated wait times on the website. While this creates value for patients, participants noted that its wider usefulness depends on broader system visibility. If a patient can see wait times across multiple services, they may make better choices about where to seek care.

This example reinforces the Government 3.0 principle: the value of a tool increases when it is connected across the system. A single hospital wait-time prediction is useful. A statewide, interoperable view of care options is more powerful.

AI Readiness Through Data Exchange Infrastructure

A whole-of-government data exchange example showed how foundational infrastructure enables future innovation. Participants described a platform supporting 1.6 billion transactions per year across 300-plus critical services, spanning all 11 Victorian Government departments and multiple agencies.

The analogy used was a “well-stocked McDonald’s kitchen” at the back end: departments should not have to hunt for data sources each time they design a service. Instead, they should be able to focus on business problems while the underlying data exchange capability provides the ingredients.

This is Government 3.0 in operational form. It is not just about connecting systems; it is about creating repeatable infrastructure that lets services be assembled faster, more consistently and with lower integration burden.

4. Challenges and Barriers

Legacy Systems Still Shape What is Possible

Several participants identified legacy technology as a major barrier. One CIO described the ongoing work of “shovelling the dirt out” by removing software that no longer fits the organisation’s needs. Others described fragmented health systems, inconsistent data semantics and platforms that do not connect with national records or consumer-facing services.

The challenge is not simply that old systems are inconvenient. Legacy systems encode old operating models. They reinforce siloed data, duplicate processes and inconsistent service experiences. Unless they are addressed, agencies may end up layering AI over brittle infrastructure.

This is particularly visible in health. Participants noted that electronic medical records, My Health Record, pharmacy information and appointment systems may each provide partial visibility, but not a seamless experience for consumers or clinicians. The result is frustration, duplication and missed opportunities for continuity of care.

Responsible AI is Not Yet an Operational Capability

The discussion made a useful distinction between governance as paperwork and governance as capability. Participants argued that responsible AI needs to become operational: something embedded into design, build, evaluation, deployment and monitoring processes.

At present, accountability, assurance and governance can be experienced as burdensome overlays. That creates resistance, especially when workforces are already stretched. However, weak governance creates its own risks: unsafe deployment, unclear accountability, poor evaluation and loss of trust.

This is especially important in high-risk contexts such as legal aid, health and justice. One participant from a risk and security background warned that the assurance layer is often the weakest point. Organisations may rely on self-assessments or vendor claims, only to discover later that controls were not operating as expected.

The pathway forward is to reframe responsible AI as value protection. It should help organisations deploy AI safely and confidently, rather than simply slow them down.

Cyber Risk and Surveillance Risk Cannot Be Treated as Afterthoughts

The strongest risk-focused contributions warned that connected systems can create new forms of vulnerability. Participants referenced recent cyber incidents affecting universities, airlines, legal services and third-party providers. The common thread was that attackers increasingly exploit the connected ecosystem, not just the core organisation.

The discussion also raised surveillance risks. The example of Flock camera networks in the United States was used to show how safety infrastructure can be repurposed for immigration enforcement or tracking people seeking abortion care across state lines. Vehicle telemetry and smart car data were also raised as examples of commercial data being made available to law enforcement.

These examples matter because Government 3.0 increases connectedness. That connectedness can improve service delivery, but it can also increase the consequences of weak governance, over-collection, unclear access rights or mission creep.

Trust therefore depends not only on whether systems work, but whether citizens believe the system will use data proportionately, lawfully and fairly.

Workforce Capacity is the Hidden Constraint

Participants repeatedly noted that the same people needed to keep services running are often the people required to design the future. This creates a structural tension between business-as-usual delivery and transformation.

Health Direct’s example captured this well. The organisation needs to maintain 24/7 service resilience, respond to health events, support national services and prepare for future AI-enabled models. A transformation office can create dedicated momentum, but subject matter expertise still sits with operational teams.

This problem is not solved by bringing in external consultants alone. If the people who understand the service are not involved in design, the innovation may not fit operational reality. But if those people are fully drawn into transformation work, service delivery can suffer.

Scaling Government 3.0 therefore requires capacity planning, not just ambition.

5. Actionable Outcomes

Immediate Actions

Start every AI initiative with the service problem.

Executives and boards should avoid mandating “use AI” as an outcome. Instead, initiatives should begin with measurable problems: reducing cost per transaction, improving time to service, reducing administrative burden, improving clinical decision support, strengthening surveillance or improving data quality.

Prioritise low-visibility, high-value back-end use cases.

Agencies should identify administrative, operational and data-processing workflows where AI or automation can reduce waste without exposing citizens to unnecessary risk. Examples include eligibility checks, document triage, staff offboarding, appointment validation and incident detection.

Build AI literacy at executive and workforce levels.

Participants stressed that many people conflate automation, machine learning, generative AI and agentic AI. This leads to poor decisions and unrealistic expectations. Executive AI literacy should focus on problem framing, risk, value, data readiness and responsible deployment.

Ask the second-order efficiency question.

If AI saves time, leaders should ask: what happens to that time? In legal aid, for example, freeing lawyers’ time could simply increase caseloads and worsen psychosocial risk. Efficiency gains need to be translated into better outcomes, not just more work.

Medium-Term Goals

Make responsible AI an operational discipline.

Responsible AI should be built into project lifecycles, not treated as a final approval gate. This means clear assurance, evaluation, monitoring, incident response, privacy review, security controls and human accountability.

Create shared value propositions for collaboration.

Collaboration improves when executives can see mutual benefit. Agencies need to define why they should give up some local control to achieve a better system outcome. This may involve shared KPIs, joint funding, common service goals or shared risk frameworks.

Strengthen data semantics, not just data pipelines.

Connecting systems is not enough if data means different things in different contexts. Government 3.0 requires attention to data models, definitions, metadata, standards and semantics. Health examples showed that data exchange without shared meaning does not create decision support.

Use universities as trusted partners for capability and experimentation.

Participants identified universities as potential partners in upskilling, experimentation and responsible innovation. They can support AI literacy, applied research, workforce development and safe testing environments.

Long-Term Vision

Build the trust architecture for connected government.

The long-term ambition is not merely more integrated systems. It is a government model where citizens, clinicians, lawyers, researchers and public servants can trust that data is used appropriately, decisions are explainable and services are connected around real journeys.

Move from pilots to production pathways.

Government needs repeatable pathways for converting successful experiments into scaled services. That requires funding models, operating models, assurance frameworks, enterprise architecture and workforce capacity.

Design for continuity of care and continuity of service.

In health, legal aid and emergency response, Government 3.0 should reduce the burden on citizens and frontline workers by ensuring the right information is available at the right time. The aim is not technological novelty. It is less friction, better decisions and safer outcomes.

Prepare for Government 4.0 by fixing Government 3.0 foundations.

Agentic and intelligent government models will only be safe and effective if identity, data, interoperability, assurance and trust are already in place. Government 3.0 is the foundation, not the side project.

6. Conclusion

The discussion made clear that Government 3.0 is both a technical and institutional challenge. Participants recognised the value of interoperable infrastructure, shared identity, connected data and AI-enabled workflows, but they were equally clear that the real barriers sit in culture, incentives, assurance, capacity and trust.

The strongest opportunities may not be the most visible ones. Back-end automation, surveillance analytics, data exchange, infection monitoring, eligibility checks and administrative workflows may deliver earlier, safer and more measurable value than highly visible citizen-facing AI tools.

The next phase of work should focus on three questions: which problems genuinely need AI, which foundational data and governance capabilities are missing, and what shared value proposition will make organisations collaborate beyond their own boundaries?

Government 3.0 will succeed when connection becomes useful, trusted and operational. Not before.