The numbers tell two different stories
The Australian Bureau of Statistics reports that 12% of businesses used artificial intelligence in 2024–25, up from 1% in 2021–22. Adoption was substantially higher in Information, Media and Telecommunications, while innovation-active large businesses led at 37%. The direction is clear: AI is moving into ordinary business activity.
Yet the same ABS release found only 10% of businesses actively collected or analysed data to inform decisions, down from 24% in 2021–22. Predictive analysis was also reported by just 6%. The measures are not identical and should not be treated as a direct scorecard, but together they expose an important risk: organisations can adopt AI tools without strengthening the information and operating practices that make their outputs useful.
Access is no longer the differentiator
A licence, an API and a polished demonstration are now relatively easy to obtain. Competitors can access similar foundation models and cloud services. What remains difficult to copy is an organisation’s proprietary context: trusted operational data, clear business definitions, established workflows and people who understand how decisions are actually made.
This changes where advantage sits. The winning question is not “Which AI product should we buy?” It is “What do we know, how reliably do we know it, and how can that knowledge improve a measurable decision or workflow?” That question naturally pulls data engineering, product management and domain expertise into the centre of AI strategy.
Move from use cases to operating cases
Many organisations maintain a list of AI ideas. Fewer define the operating conditions required for those ideas to work. A credible operating case identifies the user, the decision or task being changed, the data required, the acceptable error rate, the human accountability and the way value will be measured after launch.
That definition is an effective hiring brief. It reveals whether the next gap is an analytics engineer creating trusted data products, a product leader redesigning a workflow, a machine-learning engineer establishing evaluation and monitoring, or a domain specialist who can translate tacit knowledge into testable rules.
Build the minimum complete team
Early AI teams are often technically impressive but operationally incomplete. Model capability may be present while data ownership, integration, change and measurement are distributed across people who have no shared mandate. The result is a series of demonstrations waiting for the business to absorb them.
A minimum complete team owns a narrow outcome end to end. It combines business sponsorship, product judgement, data access, engineering, evaluation and adoption. Not every capability needs to be a new permanent hire, but every responsibility needs a named owner with enough authority and time to act.
Treat data quality as a product decision
Data quality improves fastest when connected to a specific user and consequence. “Clean the customer data” is an indefinite programme. “Ensure service agents receive complete and current account context before responding” defines which fields matter, how fresh they must be and what failure costs.
Data engineers, analytics engineers and data product leaders can then prioritise reliability in economic terms. Observability becomes a way to protect a service, not a technical accessory. Governance becomes a way to establish trust and reuse, not a catalogue built separately from delivery.
Hire for evidence of institutionalisation
As the market matures, hiring should distinguish between people who have used AI and people who have made it dependable. Ask candidates how they selected a use case, obtained access to data, established a baseline, designed evaluation, handled exceptions, won adoption and monitored the system after launch.
The strongest evidence is often found in the unglamorous middle of delivery: improving retrieval, resolving ownership, changing a workflow, reducing false confidence and deciding not to automate a judgement that needed to remain human.
The gap is still an opportunity
Australia’s adoption curve is rising, but it is not settled. Organisations that strengthen data and delivery discipline now can move beyond isolated productivity gains and build capabilities that compound. The objective is not the largest AI programme. It is a portfolio of useful systems that become more trusted as the organisation learns.
AI access will continue to broaden. Data discipline, product judgement and the ability to assemble the right team remain scarcer—and therefore more defensible.