Making AI real: Setting up for success

Love it or loathe it, AI has already started the radical evolution of government services.

We’re at an inflection point. Setting up the next generation of AI-enabled digital transformation for success is going to take some changes to the way we think about governance, strategy and implementation.

The challenges of transformation accelerated

In any government transformation there’s a tension between the strategic vision and implementation. Making it real is an artform. The tension typically lies in the disconnect between policy and delivery, unrealistic timeframes, resistance to change, restrictive technology and limited resources. With AI-enabled reform those challenges have increased. New levels of uncertainty, ethical and risk considerations, and an unprecedented acceleration in pace of technological change will make those tensions more acute. 

Digital transformation showed us how hard it is for governments to respond at pace. A lack of funds, skills, leadership and legacy systems meant that few government services are truly designed for digital. But AI isn’t waiting.

Over the last 10 years AI has grown up and is now capable of operating at a post-grad level in problem solving and reasoning. And predictions show that the pace of change will only continue to accelerate. This new capability has the potential to make today’s government services and functions more efficient, responsive and effective. But it can also help solve tomorrow’s problems by enabling new ways of thinking and working that unlock capacity to tackle entrenched challenges. The challenge is how we harness that potential.  

To set up AI projects for success, current digital transformation best practice will need to be refined. While human-centred design, systems thinking, interoperability, cyber security and agile development will remain a core part of the playbook, AI demands more agility and typical approaches to policy, strategy and implementation won’t keep up.  

Drawing insight from 9 lessons from digital transformation, this article looks at how government can navigate those challenges at a practical level.

1. New strategic agility

While digital transformation encouraged the adoption of agile implementation, government strategies remain typically waterfall, detailed plans developed over months or years. But the AI landscape is so fast-moving that those strategies will be out-dated before they are started.

We need to start developing more agile strategies.  

That doesn’t mean short-term thinking. It means iterative strategies that can respond to new insights, technologies and user needs without losing the integrity of their north star, or vision.

Building digital leadership, thinking and capability across government has been a slow journey. But we’re at an inflection point. To develop AI-enabled thinking and strategies this needs to be accelerated and embraced. AI can drive efficiencies, support human decision making, improve customer experiences and create new potential.

We need to understand the potential and be clear on what we want to achieve. 

2. A clear 2-speed approach

With AI, a two-speed reform roadmap will be a critical due to the increased levels of unknowns and risks. Fundamentally, the two-speed approach is about identifying areas where experimentation is encouraged and tested at speed, and areas of systemic reform where changes must be slower and risk minimised. A fast lane and a slow lane. 

Implementing AI in government will need an intentional and defined two-speed approach. One that clearly defines the areas where speed, transformation and risk are accepted and a slower lane for sensitive areas where AI ethical and security considerations are more significant. Start with the fast lane to accelerate learnings that can be used to de-risk the slow lane projects.

Go fast where you can, go slower where you need to. Just don’t stand still.

3. Phased, not project, funding

At a practical level, agile strategy makes detailed business cases at the outset more challenging. There’s direction and intended outcomes, but not the comprehensive detail of how to get there to support full financial scrutiny. Short-term project funding is not the answer beyond initial experimentation.

Digital transformation showed short-term project funding for large-scale infrastructure programs undermines adoption due to lack of certainty. It also showed how difficult it is to realise tangible savings.

We need to learn from that and balance prudential responsibilities with a phased and scalable funding commitment based on key milestones and long-term return on investment. 

4. Enabling pathways for local innovation

Digital transformation showed us that individuals and teams, fed up with clunky legacy systems, will create their own solutions – often without an awareness of the risks or maintenance challenges they are introducing. With AI’s plethora of easy to access tools this will increase. Rather than trying to shut the floodgates, governments need to find a way to tap into the potential and avoid uncontrolled experimentation, duplication and risk. Collaboration, recognition, enterprise pathways and mandatory security basics all form core enablers for those that want to innovate and improve through using AI.

Making responsible innovation easy and coordinated will also build much-needed AI capacity and skills across departments. 

5. Give it support to evolve

In many organisations you’d be forgiven for thinking AI equals ChatGPT or Copilot, but this limited thinking is impacting adoption and utility because it’s not automatically relevant and useful.

Digital transformation is littered with the same limited understanding and basic starting points. From departments that put their existing services online via a digital application form and didn’t go back to leverage mature digital thinking and capabilities to truly transform; to the MVP solutions that were never supported to mature to full potential and are stuck in no-man’s land.

AI, like digital, needs to mature and evolve. It’s not a tick and flick. 

6. Governance that champions a vision but understands the detail

AI is at heart experimental. It’s the opportunity to create a new reality. But the reality of AI governance needs people with the patience and technical precision of a mosaic artist who understands the purpose of each small piece and how they connect to reveal a bigger picture.

It demands a more nuanced approach to governance that can manage greater uncertainties and risk and still move forward. With the increased risk of failure, it needs to embed and normalise operating models and focus on incremental test-and-learn cycles with the controls to assess which experiments and pivots are needed to maintain overall integrity.

7. Micro innovation built for integration

AI is a contradiction. It promises big picture, blue sky, world changing impacts. But the reality of getting there is a disciplined journey of small steps. 

For example, if you’re looking to create new value or process efficiencies from variable data sources, then that journey starts with transforming one data set after another to create new value and insight. What’s the data? Where is it?  What’s it needed for?  Where’s it needed? What’s the precise action or analysis that you need AI to perform on it? Depending on the data format, function and output required there are different specialised analysers or agents to choose from.

Like Lego blocks, if they’re built for interoperability these precise micro analysis engines can work together, aggregating data, actions and analysis into a chain, much like an industrial production line. One analyses data from an application form, one applies the legislative rules, one reviews images for compliance or concerns, one compares and predicts, one aggregates the findings and injects them into decision making tools and workflows.

8. Data, data everywhere but not the stuff you need

Despite government departments being awash with information, experience shows that collecting the right data in sufficient amounts to train the analysers often takes more time and effort than initially thought.

This was an early learning for the ComplyAI portal being implemented by the Australian Government Department of Agriculture, Fisheries and Forestry’s (DAFF) Dairy, Eggs and Fish Exports Program. Although it was technically simple to enable export businesses to upload compliance information, including photos and videos for AI verification, they didn’t have a bank of photographs of non-compliant vehicles and situations to train an image-based analyser to confidently and correctly recognise potential issues.

The quality of your AI solution is determined by the quality of your data, and you need to be precise if you want accurate meaningful results. 

9. Start small and build confidence

Another key lesson from DAFF’s experiences is the value of live trials. Too often in government it’s an all-or-nothing approach where new solutions are turned on for everyone at the same time. But DAFF has worked with a small group of businesses that can see the benefits to them and are keen to collaborate to trial the new approaches and iron out the teething problems.

Go beyond co-design, find your trusted collaborators and co-create and co-test solutions in the real world.

This has been fundamental to preparing the solution for scale and to validate the anticipated improvements in compliance, productivity and faster approvals processes.  The DAFF team is now moving on to the next part of the process for improvement – learning, iterating and delivering real change piece by piece.  

Where are your next steps?

Even though it’s counterintuitive to the hype, micro improvement that’s tested and scaled incrementally towards a larger vision provides the best pathway for using AI to transform government services. But where do you start? 

In August 2025 the results of The Mandarin public sector AI survey, in collaboration with Liquid, revealed the pertinent opportunities according to your government peers. A good starting point is to decide on the problem to be solved. Do you want to:

• make internal improvements to support quicker staff decision making 
• create more automated self-service options 
• create new insights from unstructured data?

There are key characteristics, capabilities and frameworks that can help such as Liquid’s AI Leap workshop which provides a practical, structured approach to help government departments uncover where they can use AI today to unlock data, create efficiencies and improve public outcomes. But, essentially, sharing and learning from each other will be critical. And that’s where the AI survey will help. 

A version of this article was originally published in The Mandarin as part of Liquid's series looking at AI readiness in government.