In June, senior executives from Australia´s financial sector convened at a roundtable hosted by DiUS and AWS to candidly discuss their progress and challenges in scaling generative artificial intelligence across large enterprises. While interest in artificial intelligence is high and nearly every organisation is investing in experimentation, most are discovering that the largest barriers to value are not related to technology itself, but to the complexity of organisational change. Research cited by the group shows that only 1% of firms consider themselves mature in enterprise artificial intelligence adoption at scale.
Australian financial services organisations are scattered across the adoption curve: some have yet to begin, others are trialling copilots, chatbots, or sandboxes, while a handful run mature targeted tools or internal platforms. However, even the most advanced teams continue to grapple with ´the messy middle´—the challenge of moving from isolated proofs of concept to repeatable, scalable capabilities that are fully embedded into governance, delivery, and daily business practices. Common obstacles include difficulties in rolling out generative platforms across the enterprise, lack of developer trust in copilot outputs, and compliance barriers in tightly regulated environments. DiUS advises that success is found by enabling artificial intelligence throughout the entire delivery lifecycle, building capability in tandem with governance, data, and engineering practices, and maintaining focused, outcome-driven priorities over exhaustive strategies.
Articulating the value proposition of generative artificial intelligence remains a core struggle. Teams often default to seeing artificial intelligence as a general capability promising smarter, faster operations, rather than defining direct business outcomes. This lack of precise metrics complicates the case for investment, particularly as infrastructure costs mount. DiUS recommends anchoring initiatives in measurable improvements, using frameworks like desirability, viability, and feasibility (DVF) to prioritise opportunities. Organisations have seen tangible results, such as reducing time-to-insight by 40% or enabling staff to process complex policy documents much faster.
Technical and structural hurdles persist. Leaders highlighted issues such as data fragmentation, latency, privacy, security, and rapidly evolving systems, which breed caution rather than outright resistance. Data quality emerges as the most significant bottleneck—many organisations lack reliable, integrated, and well-governed data sources, which hampers artificial intelligence reasoning and trustworthy outputs. The move toward more autonomous ´agentic´ artificial intelligence systems also demands reusable architecture, orchestration patterns, and evolved trust frameworks, along with new team skills and governance sophistication.
Change management is identified as a crucial multiplier. Technology alone cannot deliver transformation; the human transition—training, support, and cultural adaptation—remains severely underestimated. Successful financial enterprises are treating generative artificial intelligence as a long-term capability rather than a stand-alone project, investing in shared platforms, communities of practice, and iterative skill development. Ultimately, the pathway to scalable, valuable artificial intelligence in Australia’s financial sector depends on clarity of goals, flexible architectural foundations, and a proactive approach to organisational change.
