Australian firms not ready for AI at scale, study warns

Interactive has published research showing few Australian organisations are ready to deploy artificial intelligence at scale. Only 8% of data and AI leaders said their data was optimised for that task.

The study also found weak returns from AI agents, with only 15% of surveyed Chief Information Officers and Chief Data and AI Officers saying results met or exceeded expectations.

Research compiled by ADAPT Research & Advisory for Interactive's Unstoppable AI Blueprint suggests the main barriers extend well beyond core technology systems. Among Chief Information Officers asked which areas were least mature for scaling AI, data foundations and quality ranked highest at 40%, followed by workforce capability and adoption and process design and operational readiness, both at 38%. Governance and risk management came next at 36%, ahead of value realisation and measurement at 34%.

Technology and platforms ranked last at 25%, suggesting many leaders see organisational readiness, data quality and operational control as bigger obstacles than infrastructure alone.

Lizzy Jones, Head of Data and AI at Interactive, said the quality of the underlying environment determines whether AI projects produce meaningful outcomes.

"AI doesn't sit above the rest of the technology estate. It amplifies whatever is already there, strengths and weaknesses alike. What matters is the data a use case depends on, who can access it, where the workload runs and who owns the outcome," Jones said.

Agent concerns

The report places particular emphasis on agentic AI, in which systems can take actions rather than simply produce responses. It found 71% of Chief Information Officers said their organisation was not prepared, or only slightly prepared, to manage an agentic workforce.

Only 9% said they had enterprise-wide controls over what AI agents can access and act on. Returns from agentic AI were also low, with only 13% of Chief Information Officers reporting outcomes that met or exceeded expectations.

Chris Pallister, Head of Managed Security Services at Interactive, said operational risks rise quickly when autonomous systems are introduced without clear safeguards.

"AI agents know a huge amount and reason incredibly fast, but they know nothing about your business. Combine that with access most organisations can't even fully see, and small mistakes become big problems very quickly," Pallister said.

"Before an AI agent goes anywhere near production, leaders need to know three things: what it can access and change, who owns it, and how quickly they can switch it off. An agent that breaks is easy to spot. The real danger is one that keeps running, quietly getting things wrong, and nobody notices," he said.

Infrastructure pressure

The research suggests Australian companies are reassessing where AI workloads should run as they face cost pressure and gaps in infrastructure readiness. Only 9% of infrastructure leaders rated their environment as highly or fully ready for AI workloads, while 77% said they were under active cost pressure.

FinOps maturity remains limited, with just 14% reporting mature practices. At the same time, 28% said they planned to repatriate some public cloud workloads, up from 10% in 2020, while 64% said onshore AI inference was very or critically important for future small language models.

Those figures point to a more cautious approach to deployment as businesses balance spending, operational control and data location requirements. The survey suggests infrastructure choices are becoming more deliberate as AI use expands across organisations.

Internal rollout

Interactive also disclosed examples from its own internal use of AI. Enterprise AI agents are helping it onboard vendors and suppliers about 20 times faster, while AI-assisted triage and investigation in its Security Operations Centres have cut average threat detection and response times from 40 minutes to seven.

Jones said those outcomes depended as much on staff adoption as on the systems themselves.

"Our biggest lesson from Customer Zero is that AI adoption is a people challenge before it is a technology challenge. Adoption moved from curiosity to habit when people found practical use cases that saved them time or removed friction from their day. That only happened because we prepared people before we rolled out the tools, not after," she said.

The survey covered senior Australian executives across data and AI, information technology, and cloud and infrastructure functions. The results add to a growing body of evidence that board-level enthusiasm for AI is outpacing the practical work needed on data quality, governance, cost control and workforce readiness.

For many organisations, the study's message is that scaling AI depends less on adding another model and more on fixing the conditions around it. Only 8% of leaders said their data was optimised for AI at scale.