Australia's AI Returns Are Rising Faster Than Its Guardrails

There is a number in the latest SAP research that will please a lot of Australian boards, and a second number sitting right next to it that should give them pause.

Australian companies now forecast AI return on investment of 19 percent this year, up from 15 percent last year, with that figure expected to reach 37 percent within two years (Source: SAP and Oxford Economics, 'Value of AI Report 2026'). The research surveyed more than 2,600 business leaders across 13 countries, and AI already supports a meaningful share of the working day here. As the report puts it, AI now supports 29 percent of tasks in the average Australian business, up from 25 percent a year earlier, with leaders expecting that to reach 48 percent within two years (Source: SAP and Oxford Economics, 'Value of AI Report 2026').

So returns are climbing, spend is climbing, and AI is doing more of the work. That is the good news. The catch is the number most of the coverage skipped past.

The gap the ROI headline hides

On governance readiness, Australia is behind. Only 22 percent of Australian organisations said they were mostly or fully ready in AI governance, compared with 33 percent globally (Source: SAP and Oxford Economics, 'Value of AI Report 2026'). In other words, we are chasing bigger returns faster than we are building the controls to manage them.

That matters because ROI and governance are not separate tracks. The value you keep from AI is the value your controls let you hold on to. A tool that saves an hour a day but quietly ships a wrong price to a customer, or leaks data it should not touch, does not sit neatly on the return side of the ledger. It moves to the cost side, often with interest.

When adoption outruns oversight

The survey is blunt about where this is already biting. Some 42 percent of respondents said they were deploying AI agents faster than they could standardise and govern them, while 54 percent said employees were increasingly accepting AI outputs without enough scrutiny (Source: SAP and Oxford Economics, 'Value of AI Report 2026').

And it is not hypothetical. The findings showed 43 percent of Australian organisations do not have human-in-the-loop processes for agentic workflows, and 49 percent said AI agents had already taken incorrect actions during pilots or deployment, usually leading to rework and delays (Source: SAP and Oxford Economics, 'Value of AI Report 2026'). Rework and delays are exactly the kind of hidden tax that eats into the return you were counting on.

SAP's own local leadership framed it as a work in progress. Angela Colantuono, president and managing director of SAP Australia and New Zealand, described the results as an incomplete report card: improving, but not yet working to potential (Source: SecurityBrief Australia, 'Australia lags on AI governance as adoption races ahead', 2026).

What governance readiness actually needs at pace

The timing is the part leaders should not miss. On 15 July 2026 the Prime Minister used a keynote titled 'AI in Australia's interests' to set out a shift from voluntary standards toward a mandatory, whole-of-government framework, with a dedicated Office of AI, National Cabinet consideration in August and legislation expected in early 2027 (Source: White & Case, 'Australian AI Update: Australia changes course', 2026). So the oversight gap is widening at the same moment the rules are firming up. You do not want to be closing both at once under deadline pressure.

The good news is that governance at pace does not mean slowing adoption. It means a few concrete habits that scale alongside it:

• Name an accountable owner. Give every AI use case a clear executive owner and someone responsible for approving, running and monitoring it. Ambiguity here is where risk hides.
• Keep a live inventory. Maintain an AI register so you know what is running, where, on what data, and who signed off. You cannot govern what you cannot see, and shadow use grows fast.
• Put a human in the loop where it counts. Match the level of oversight to the stakes of the decision. Agentic workflows touching money, customers or personal data need a checkpoint, not blind trust in the output.
• Work to a standard you can carry forward. The National AI Centre's Guidance for AI Adoption sets out six essential practices and evolves the earlier Voluntary AI Safety Standard, and it aligns with ISO/IEC 42001 and the NIST AI Risk Management Framework (Source: Department of Industry, Science and Resources, 'Voluntary AI Safety Standard', 2026). Building to that now means you are already most of the way to whatever lands in 2027.

None of this is exotic. It is the difference between AI you can defend in front of a board, a customer or a regulator, and AI you are quietly hoping nobody asks about.

The returns are real. The task for Australian leaders is to make sure the oversight keeps pace with the ambition, so the 37 percent you are forecasting is a number you actually get to keep.

The right next steps for you depend on where you are right now. Our AI Maturity Assessment is an easy place to start. After that, let's have a chat about your next steps.