Half of Australian businesses report AI agent errors, yet 62% say they’re satisfied with returns

Sixty-two percent of Australian businesses say they are satisfied with their AI returns. In the same dataset, 49% report AI agents have already taken incorrect actions in pilots or production, and only 22% consider themselves governance-ready. A majority of the leaders surveyed say employees increasingly accept AI outputs without sufficient question.

Those findings sit together in SAP’s Value of AI Report 2026, released this week and conducted by Oxford Economics across more than 2,600 business leaders in 13 countries. They land in the same week the Prime Minister announced mandatory Australian Standards for AI, and alongside a body of SAPinsider benchmark research that points to similar gaps inside SAP landscapes globally.

Angela Colantuono, President and Managing Director of SAP Australia and New Zealand, framed the situation as a report card. “Think of this as Australia’s AI school report: improving, but still not working to its potential.” She warned that with multiple new AI obligations coming into force before the end of the year, organisations need to strengthen data foundations, governance frameworks and leadership structures. “Otherwise, we risk building the rails for an AI economy without being ready to run on them.”

The biggest bets are going to the least governed capability

The scale of spending makes the governance gap harder to set aside. Australian organisations expect to invest approximately AU$35.5 million on AI this year, up from AU$27.5 million, with a further 44% increase forecast over the next two years. ROI expectations have climbed from 15% to 19%, heading toward 37%.

The fastest-growing slice of that investment is agentic AI, where expected ROI has more than quadrupled from AU$4.4 million to AU$21.2 million over two years. That concentration of expected returns sits against SAP’s finding that only 19% of Australian organisations are scaling or leading on agentic AI, trailing the global average of 24%. It also sits against the governance numbers: 42% of businesses concede they are deploying agents faster than they can standardise or govern them, rising to 68% when those who are unsure are included. Forty-three percent have no human-in-the-loop process for agentic workflows. Australia’s governance readiness of 22% trails the global average of 33%.

Professor Toby Walsh, Scientia Professor of Artificial Intelligence at UNSW, put the underlying risk plainly: “Artificial intelligence is only valuable if people know when to trust it, and when to question it. Good governance shouldn’t be viewed as slowing innovation. It’s what allows organisations, employees and the broader community to adopt AI with confidence.”

The same gap, viewed from inside SAP landscapes

SAPinsider’s 2026 benchmark research, surveying the SAP practitioner community between late 2025 and Q1 2026, surfaces a version of the same tension at the platform level.

In its Technology Leader’s Strategic Agenda for 2026 report, only 16% of technology leaders say they use AI in more than a limited manner in SAP-related systems, even as 40% name intelligent automation in core ERP as a planned use case and another 40% name predictive analytics. Deploying Joule or embedded AI ranks third among planned SAP initiatives but fifth among actual planned investments. The two biggest challenges technology leaders identified were aligning IT and business stakeholders on priorities (51%) and managing cost pressures while funding transformation (49%).

That investment sequencing, where AI slips down the priority list once budgets are committed and the money flows to BTP, analytics and infrastructure, may reflect a recognition among technology leaders that the layer underneath the AI has yet to be built.

SAP’s Australian research and SAPinsider’s global benchmarks converge on this point. SAP’s report found Australian leaders rank integrated data systems (64%) and data quality (51%) as the biggest enablers of AI readiness, yet the share of businesses describing themselves as data-ready for AI fell from last year, and 73% report challenges with poor data quality. SAPinsider’s Enterprise Data and Analytics in the Era of AI report puts transformational data and analytics maturity at 11% globally, with more than 40% of organisations holding only partially integrated data across SAP and non-SAP systems.

SAPinsider’s SAP Business Data Cloud research quantifies the architecture gap more precisely. Three percent of organisations have the unified, governed data layer the platform is designed to accelerate. Thirty-eight percent remain in siloed or ad hoc integration states. Among organisations evaluating or deploying BDC, 38% report no measurable outcomes yet. Only 12% have the automated governance required to support AI-driven workloads. And on the agentic AI ambition specifically, BDC research shows agentic workflows entering early production roadmaps at 24% of organisations, while SAPinsider’s data analytics research found Data Leaders are twice as likely as Data Adopters to be running agentic AI at all. The maturity floor required to deploy agents productively is one most organisations have not yet reached.

The S/4HANA race and the AI it leads to

SAPinsider’s ERP Migration and Transformation 2026 report adds a further complication. SAP’s announcements on AI have become the single largest external factor reshaping ERP strategy, cited by 43% of respondents, up from 15% a year ago and now ahead of the 2027 end of mainstream maintenance deadline at 39%. Customers are, in growing numbers, framing their S/4HANA migration as an AI migration.

The challenge is what that migration delivers in practice. The report notes that the AI capabilities available to every S/4HANA customer are iterative, not generative, and that accessing Joule requires an SAP Cloud ERP contract many organisations have yet to secure. Meanwhile 35% of respondents say they plan to complete the switch to S/4HANA before the end of 2026, more than double last year’s figure. SAPinsider’s own assessment of that timeline: “Unless these deployments are already underway the timeline for transition may be somewhat optimistic,” adding that “this number may end up representing more of an intention than a reality.” With an estimated 20,000 to 25,000 SAP ERP customers still to license S/4HANA, the report warns of “a significant resource bottleneck over the next two to three years.”

Canberra sets a deadline

The governance gap SAP’s research documents and SAPinsider’s benchmarks mirror is about to meet a regulatory timeline. Prime Minister Anthony Albanese’s speech at the University of Sydney on 15 July announced mandatory Australian Standards for AI, to be taken to National Cabinet next month with legislation targeted for early next year. A new Office of AI will sit inside the Department of the Prime Minister and Cabinet to coordinate the framework across government.

The standards cover data centre energy, water and location obligations, and legislate copyright control for Australian creators and media. But the broader signal matters more to SAP customers than the specific provisions. Albanese framed the announcement as a sovereignty argument rather than a precautionary one: “Our great country can be much more than a data warehouse for AI products made overseas.” He pointed to the under-16 social media ban as evidence that Australia setting a standard early led more than 20 nations to follow, and made clear the government intends to lead rather than accommodate.

SAP’s research shows the regulatory obligations around AI use are already stacking up. Ninety-nine percent of Australian organisations operate under some form of sovereign AI framework or requirement, and 76% cite data residency constraints limiting model choice. Mandatory national standards will add a further layer. By SAP’s own numbers, those standards will arrive in an environment where 42% of businesses are deploying agents faster than they can govern them and nearly half have experienced agent errors.

Colantuono pointed to the urgency directly: “With multiple new AI obligations coming into play for Australia between now and the end of the year, driven by local and international regulation, organisations need to strengthen their data foundations, governance frameworks and leadership structures.”

What this means for Mastering SAP insiders

Two research programs and the federal government have now converged on the same point: that AI investment, and agentic AI investment in particular, is outpacing the data architecture and governance required to make it productive and compliant.

The SAPinsider data maturity research offers the clearest benchmark for where that investment should go. Organisations at the top maturity tier are twice as likely to run agentic AI as the middle tier, more than twice as likely to use natural language querying for BI, and seven times more likely to use synthetic data or digital twin modelling. Forty-two percent of these Data Leaders plan to significantly increase data and analytics investment over the next 12 months, compared with 13% of Data Adopters planning the same.

The budget signal in SAPinsider’s technology leader research, where AI drops from third priority to fifth when actual investment decisions are made and the money flows to BTP, analytics and infrastructure, suggests many technology leaders already understand this sequencing in practice even where their strategy documents do not yet reflect it.

SAP’s Australian data, SAPinsider’s global benchmarks, and the Prime Minister’s legislative timetable all point in the same direction. Colantuono’s warning about building rails without being ready to run on them now has a regulatory deadline attached. The organisations most exposed are not those moving slowly on AI, but those scaling agentic deployments without the data foundations, governance processes, and human oversight that SAP’s own research describes as prerequisites.