← Field Notes
·6 August 2026·4 min read

Australia's AI ROI Problem Isn't the AI — It's the Data

SAP's 2026 report: Australian businesses expect AI ROI to hit 37%, but 73% say poor data quality is throttling it and only 22% feel ready on governance.

Australian businesses expect their AI investments to return 37 per cent within two years — nearly double this year's 19 per cent. That is the optimistic half of SAP's Value of AI Report 2026, published on 20 July. The other half is the reason most won't get there, and it has nothing to do with the AI.

Seventy-three per cent of Australian businesses in the report say poor data quality is holding their AI back. Only 22 per cent rate themselves ready on AI governance, against 33 per cent globally. And 42 per cent admit they are switching on AI agents faster than they can govern them. The ambition is real. The foundation underneath it is not.

SAP's report, run by Oxford Economics across more than 2,600 business leaders in 13 countries, is one of the more grounded reads on where AI value is actually landing. In the Australian cut, AI now supports 29 per cent of tasks in the average business, up from 25 per cent a year ago. Leaders expect that to reach 48 per cent within two years. The tools are in. The spend is climbing.

But the report is blunt about what is throttling the return, and it is two unglamorous things. First, data: 73 per cent report challenges with poor data quality, and 64 per cent name connected, integrated data systems as the single most important enabler of AI value. Second, governance: beyond the 22 per cent who feel ready, 43 per cent have no human-in-the-loop process for the AI agents they are already running.

73%

Cite poor data quality as a barrier

Top drag on AI value

22%

Feel ready on AI governance

33% globally

43%

Run AI agents with no human check

No human-in-the-loop

Here is the part that matters for a ten-person business. An AI agent can only reason over the records you give it. Point one at a trades business whose job history lives half in a job-management app and half in a glovebox full of dockets, and it will produce confident, wrong answers — because it is working from half the picture. Point one at a professional firm whose client data is scattered across email threads, spreadsheets and a legacy practice-management system that cannot export, and the agent spends its intelligence stitching together what should already be one clean record.

This is why the 64 per cent figure — integrated data as the top enabler — matters more than the ROI headline. The return on AI is not set by how clever the model is. It is set by how clean and connected your data is before the model ever touches it. That is a Cost Intelligence problem before it is an AI problem, and much of it is fixable without buying a single new tool. If you are not sure whether your data is ready for the AI you are already paying for, a short conversation will tell you where the gaps are.

Want to know what this means for your business? We'll map it in one call.

Book a call →

The second finding — 42 per cent deploying agents faster than they govern them — sounds like a big-company problem. It is not. The smaller the business, the more likely a single ungoverned agent is drafting quotes, answering customer emails or summarising advice with no one checking it before it goes out. When it gets something wrong, the business owns the mistake. We have written before about who is on the hook when an AI tool gets it wrong under Australian Consumer Law — the short answer is you, not the vendor.

Human-in-the-loop — a person signing off on anything an agent produces that reaches a customer — is the cheapest insurance in AI. The 43 per cent running agents without it are not saving time; they are deferring a problem. It is the same pattern we keep seeing: Australia ranks near the top of the world for AI readiness and near the bottom for putting it to work. The gap is not capability. It is the boring foundational work between buying AI and getting value from it.

Expected return on AI investment (Australia)

This year

19%

Reported ROI

In two years

37%

If the foundation holds

Before adding another AI subscription, run a one-page check on the foundation. Three questions. Where does your business data actually live, and can your core systems talk to each other? Is that data clean enough that you would trust a decision made on it? And for anything an AI produces that a customer sees, does a human sign off first? If you cannot answer those cleanly, that is where the next dollar goes — not on more AI. Fixing the data and the controls is what turns a 19 per cent return into the 37 per cent the report says is on the table.

Key takeaways

Australian businesses expect AI ROI to nearly double to 37% within two years, but SAP's Value of AI Report 2026 (20 July) finds the returns are throttled by weak foundations, not weak technology.
73% of Australian businesses report poor data quality holding their AI back, and 64% name integrated, connected data systems as the single most important enabler of AI value.
Only 22% feel ready on AI governance (33% globally), and 43% run AI agents with no human-in-the-loop — an unmanaged liability, since the business, not the vendor, owns the mistakes.
The fix is foundational and cheap: clean, connected data and a human sign-off on anything customer-facing an agent produces — before spending more on AI.

Common questions

Why isn't AI delivering ROI for my business?

SAP's Value of AI Report 2026 points to foundations, not the technology: 73 per cent of Australian businesses cite poor data quality as the main drag, and 64 per cent say connected, integrated data systems are the single most important enabler of AI value. An AI tool can only reason over the data you give it — if that data is scattered or unreliable, the return is capped before the model starts.

What is human-in-the-loop, and does my small business need it?

Human-in-the-loop means a person reviews and signs off on anything an AI agent produces before it reaches a customer. The report found 43 per cent of Australian businesses run AI agents without it. For a small business it is cheap insurance: under Australian Consumer Law the business, not the AI vendor, owns the mistake when an automated output is wrong.

Do I need clean data before adopting AI?

Largely, yes. The report found data quality is the top barrier to AI value for Australian businesses. You don't need perfection, but your core systems should be able to talk to each other and the records should be reliable enough to trust a decision made on them. Fixing that is often a bigger lever on AI returns than buying another AI tool.

Sources

SAP — Australia's AI report card: “Improving… but could do better” (Value of AI Report 2026, 20 July 2026)

Assumptions & methodology
  1. All figures are from SAP's Value of AI Report 2026, released 20 July 2026 and conducted by Oxford Economics across more than 2,600 business leaders in 13 countries. The percentages cited here are the Australian findings as published in SAP's Australian release; the 22 per cent governance-readiness figure is explicitly contrasted with 33 per cent globally in that release.
  2. The 19 per cent and 37 per cent figures are the expected return on AI investment reported for this year and anticipated within two years respectively, in the Australian cut of the report.
  3. The report's investment-dollar figures reflect an average across large and mid-sized organisations and are not representative of small-business budgets, so they are not cited here. The data-quality, governance and ROI ratios generalise across business size and are the focus of this note.

Related notes

ShareLinkedInEmail

Field Notes are general commentary on AI trends for Australian businesses. They don’t constitute professional advice. Talk to your accountant, lawyer, or IT adviser before acting on anything specific to your situation — or talk to us if you want help working out where AI fits.

Don't miss the next one

Get each new Field Note in your inbox as it publishes — short, practical AI intelligence for business owners.

Not sure your data is ready for the AI you're paying for?

A short conversation will map where your data and controls stand — and the one fix that will do most to lift the return on the AI you already run. Book a call to talk it through.

Book a call →