From pilots to platforms: How government can scale AI safely
Governments don’t need more AI pilots. Safe, scalable and flexible AI platforms are the way forward.
Fiona Armstrong

AI is everywhere – but mostly in the form of hundreds, even thousands of small pilots. Chatbots here, dashboards there, scribes in consulting rooms, predictive models tucked away in data teams. Each pilot may be useful, but together they create a system that is fragmented, duplicative and unsustainable.

Pilots have their place. They are low-risk ways to test ideas, explore unfamiliar technologies and build confidence. But as a long-term strategy, they send the wrong signals and fail to deliver systemic value.

The Mandarin/Liquid national survey of public servants’ attitudes to AI captured workers’ frustration at the “stop-start nature of pilots”. It also revealed another source of frustration; too often, governments started by rolling out generic chat tools that were not right for the problems they wanted to solve – such as eligibility checks, triaging cases, or surfacing insights. All the while, chief information officers (CIOs) and other government leaders were navigating how to manage the risk of letting a thousand blossoms bloom.

This misalignment risks stalling adoption momentum if tools remain ad hoc and uncoordinated.

Now is the time to take a new approach: AI as a platform
AI platforms can provide safe, secure, flexible and scalable tools that are easily configured to create fit-for-purpose solutions. AI platforms are different to applications like Copilot or ChatGPT, or the new AI module inside your current software. They can integrate or leverage those and many more AI/ML tools to create connected solutions that are fit-for-purpose.

AI platforms also provide an easy way to leverage the learnings and efforts of others in government. Rather than scattered and wasted investments and teams duplicating things that have been done before, an AI platform can accelerate change and scale with usage.

This approach is seen in leading AI-adopter governments around the world:

In Singapore, the government established a reusable AI/ML stack with AI verification for agencies, avoiding duplicate buildouts and speeding up safe reuse.
The US Veterans Affairs Department confronted this problem directly, with 200+ AI pilots. Few had scaling potential because they lacked integration and governance. Their solution was to establish a central AI portfolio and platform to replace fragmentation with intentional scaling.
AI platforms can also accelerate improvements to citizen experiences through consistency and trust:

In Finland, Aurora AI is a national AI network and operating model designed to link public/private services around citizen life events where ethics and citizen trust are foundational.
In Estonia, Bürokratt AI is a state platform that provides a single, united channel into public services, with shared components and open standards replacing 130+ AI tools.
While pilots create learnings, it’s platforms that enable transformation
Language matters. Platforms say: “This is a core part of the way we do things around here. This is a safe space. Don’t worry about the tech, focus on fixing the problems that need solving.” As such, they support the cultural shift that’s needed to build AI adoption. Pilots, on the other hand, say: “This is just a short-term experiment — don’t invest. This can’t be scaled.”

In the UK, an AI platform meant the NHS could take an AI stroke triage tool that began in a single hospital and scale to all 107 stroke centres in the country, saving critical time in treatment and improving recovery rates for patients. Scale matters where equality matters.

If we want to see AI adoption in government at scale, we need to use AI platforms to create minimum viable services that we test and iterate on.

AI should never be “set and forget”. By its nature, AI learns from data that evolves, operates in dynamic contexts and carries shifting risks. That’s why scaling safely is about iterative learning and adaptation at the platform level, not endless one-off pilots.

In Australia, an early AI platform adopter, the Department of Agriculture, Fisheries and Forestry (DAFF), is proving the value of an iterative approach — where the AI platform allows continual learnings to be fed back in to make the process and outcomes better, then leveraged for the next service to be transformed.

Each of these examples reveals the core characteristics of an AI platform approach:

Shared infrastructure
Reusable components
Integrated governance
Flexibility across use cases
Citizen-centred design
They also highlight the need for leadership commitment, knowledge sharing, skills development and openness to rethinking how government operates. An AI platform works best with strong governance, clear prioritisation and built-in adaptability.

As DAFF regulator Shayne Daniels explains: “Carefully building trust is critical. If we solely relied on AI, then we would fail. Any good system requires human verification that sits over the top of it. But the beauty of our Unearth AI platform is that the information we’re working with is more regular — sometimes even real-time — and the automated compliance checks mean our team can focus on more complex issues. It’s helping us, and the businesses we work with, to create better outcomes.”

Pilots have taught us enough. Now the task is safe, coordinated scaling
The Mandarin/Liquid “Let’s get real about AI” survey respondents want investment focused on reducing administrative burden (42%) and supporting decision-making with insights (20%), far more than speculative or experimental use cases. That aligns with starting in low-risk, high-friction areas. Rather than scattering across disconnected pilots, adopting an AI platform will enable governments to realise that potential quicker and build the confidence and skills for the more transformative opportunities ahead.

Scaling isn’t about having more AI – it’s about having better, safer, more trusted AI. It’s about moving from snapshots to continuous assurance, from fragmented pilots to integrated platforms, from experiments to systems.

This is the inflection point. The pain points are well understood. The policy scaffolding is in place. The workforce is ready to lead. The challenge now is to shift from aspiration to execution.

From pilots to platforms: How government can scale AI safely

Sign up to our newsletter.