Big bang AI solutions lead to messy outcomes: here’s how you avoid them

AI is projected to contribute up to around $235 billion to Australia’s economy, according to the Australian Academy of Technological Sciences and Engineering. Investment in the sector is accelerating, with Microsoft recently committing $25 billion to AI infrastructure, security and skills in Australia. The federal government has also committed billions towards AI, digital capability and innovation through the National AI Plan and related initiatives.

But ambition must be balanced with reality. Organisations are eager to deploy AI but often lack the readiness to do so. Poor data integration, fragmented architectures, inconsistent meaning, and talent shortages create barriers that prevent AI initiatives from moving successfully from pilot projects to full production.

Legacy systems, both technological and cultural, can limit the impact of AI. When leadership and teams resist change, even successful pilots struggle to scale.

As agentic AI opens new possibilities, organisations need to overcome the inertia they create. This requires rethinking workflows and redefining responsibilities across teams, not just introducing new technology.

Set objectives and invest in the right building blocks

Effective deployment begins with understanding use cases, business context and the maturity of different AI capabilities. Select use cases carefully, not just ‘low-hanging fruit’. Prioritise based on return on investment, adoption potential or strategic value. Since AI maturity varies by domain, organisations should focus on applications that are feasible today.

Getting data fundamentals right matters: AI depends on data, and fragmented environments hinder outcomes. Waiting for perfect architecture, governance or semantics stalls progress. Instead, understand your data estate, make it visible, and design AI within those constraints while incrementally improving them.

For example, when South Australia’s Department for Education set out to use advanced analytics, it found that existing student data practices wouldn’t scale due to fragmented access across multiple systems and custodians. Instead of resolving every upstream constraint, they focused on understanding these limitations and designing a platform to operate within them, improving access to accurate, timely data, even with external dependencies.

Siloed ownership and access are real constraints. Progress depends on collaboration, aligning architecture to data contracts, and incrementally strengthening the underlying data estate while delivering value.

Developing an internal framework isn’t about creating more structure, it’s about enabling teams to operate with shared context, consistent meaning and clear boundaries. Without that, scaling AI amplifies fragmentation.

Many organisations scale too early, before aligning on how data is defined, accessed and used, resulting in disconnected initiatives, duplicated logic and inconsistent outcomes.

Instead of defining everything upfront, organisations should focus on a few high-value use cases. This includes being explicit about who the users are, what decisions are being supported, and how outcomes will be measured — forming the foundation for how the architecture evolves.

Then, integration is about ensuring AI operates consistently across existing systems. This is where semantics, shared definitions and governed access start to matter. Without them, APIs and models remain technically connected but operationally disconnected.

Sustained AI adoption requires governance, not as a blocker, but to clarify acceptable, secure and compliant practices. In regulated environments this balance is critical: too little control creates risk, while too much stops progress altogether.

The organisations making progress are not waiting for a perfect framework. They are building one iteratively, grounded in real use, and strengthening it as AI adoption grows.

Focus on culture

Australia’s National AI Plan encourages the responsible development and deployment of AI systems. While a good starting point, it fails to emphasise the important role of culture and how teams actually work, or are meant to work.

AI does not normally fail for lack of ideas. Failure often stems from disconnected processes, unclear guardrails and differing interpretations of data and outcomes among teams. Culture significantly influences whether AI scales effectively or becomes fragmented.

Given AI cuts across data, engineering, risk and business teams, without a shared way of working, and a common understanding of semantics, teams tend to separate in their approaches, causing initiatives to lose momentum.

Enablement isn’t just about training people on tools: it involves providing them the right context, boundaries and confidence to act. Clear frameworks, access to governed data, consistent meaning and well-defined guardrails allow teams to move without constantly second-guessing what is acceptable or correct.

For one banking customer, capability uplift has been as important as the platform itself. Teams have been supported in understanding and using AI, not just technically, but also in how it fits into their day-to-day decision-making processes.

The next step is shifting from centralised delivery to guided self-service. When done properly, this allows authorised users to access and work with data directly, focusing on higher-value analysis rather than repetitive data requests. Done poorly, it simply creates more inconsistency.

This is the balance organisations need to strike. Broad AI adoption requires people at all levels to engage, but within a framework that keeps meaning, access and outcomes aligned.

Not a one-time fix

AI is not a silver bullet: it amplifies the current operating model, good or bad. Deploying it into fragmented environments often scales that fragmentation.

For AI to deliver real value, it needs to sit within a coherent framework that provides oversight, clear boundaries and processes to manage issues. Progress depends on continuous iteration, with each use case improving capabilities and the underlying foundations.

While cloud strategies can accelerate AI initiatives, they are only part of the story. Moving and processing large amounts of data at scale doesn’t make it more meaningful. Without consistent definitions and a clear structure, scale only amplifies inconsistency.

What enables AI to adapt and operate reliably is a data architecture grounded in shared meaning. Standardised definitions and, increasingly, formalised ontology allow AI systems to interpret data consistently across different domains, avoiding fragmented outcomes.

Australia is well-positioned to capitalise on the AI opportunity. The next step is to focus on governance, efficient operations and talent to pivot and achieve meaningful results for organisations at every level. *Etienne Oosthuysen is Chief Technology Officer at Synechron Australia with expertise across data, AI, cloud and digital engineering. He operates at the intersection of technology strategy, client engagement and commercial outcomes, connecting capabilities across enterprise platforms to address complex challenges and shape cohesive solutions.

Top image credit: iStock.com/BlackJack3D