AI ROI Jumps to 21% as Agentic Systems Drive Enterprise Returns

AI investments deliver measurable returns despite persistent challenges

Businesses worldwide are extracting significantly higher returns from artificial intelligence investments, according to new research from SAP and Oxford Economics that surveyed 2,600 leaders across 13 countries. Companies now report 21% ROI on AI spending—up from 16% last year—and project returns will climb to 38% within two years.

The average global business now spends $28 million annually on AI, a modest increase from $26.7 million in 2025. That investment currently generates $6.3 million in returns, with expectations to reach $15.9 million by 2028. Regional patterns show significant variation: AI spending surged in Brazil, the UK, Australia, and Germany, while investment declined in China and India.

Agentic AI systems—autonomous agents that can execute complex workflows with minimal human intervention—account for much of the optimism. Expected ROI from agentic implementations is projected to reach $17.6 million within two years, more than quadrupling from last year's $4.3 million estimate. Eighty-three percent of surveyed businesses view agentic AI as having moderate to very high transformational potential.

Why it matters

The research reveals a critical inflection point: AI has moved beyond pilot programs to deliver measurable business value, yet most organizations lack the governance infrastructure to scale safely. As agentic systems gain autonomy, the gap between deployment speed and oversight capability creates enterprise risk that could undermine the substantial returns now materializing.

Governance and data quality emerge as critical bottlenecks

Despite growing returns, fundamental challenges persist. Data quality remains the primary obstacle, with 73% of companies reporting incomplete data issues—an increase from the previous year. Poor data quality has operational consequences: 79% of businesses experience rework, delays, or backlogs due to low-quality AI outputs.

Governance capabilities lag dangerously behind adoption rates. Only 12% of businesses report their skills, processes, and frameworks are fully ready to govern AI effectively. The governance deficit becomes more acute with agentic systems: 38% of companies lack human-in-the-loop processes for agentic workflows, 37% have no permission and access controls for agents, and just 44% maintain a registry of deployed agents. Sixty-nine percent of businesses either agree or remain uncertain whether they're deploying agents faster than they can govern them.

Workforce readiness also presents challenges. Seventy-eight percent of organizations report that upskilling efforts aren't keeping pace with AI tool evolution, while shadow AI use—employees deploying unauthorized AI tools—occurs at least occasionally in 69% of companies.

Strategic approaches still uncommon

AI now supports 30% of all tasks in the average business, a figure expected to reach 48% within two years. Yet strategic, enterprise-wide AI investment accounts for just 17% of spending—nearly double last year's level but still far behind piecemeal approaches, which represent 41% of implementations.

Leadership structures remain underdeveloped: fewer than half of companies have a dedicated AI leader (46%), clear AI development frameworks (52%), or training on AI capabilities and risks (41%).

"AI has moved from experiment to execution, and that's beginning to show real returns," said Sean Kask, SAP's Chief AI Strategy Officer. "But there's still a long way to go. Because AI that lacks context—whether that's processes, data, or governance—at best creates activity without outcomes and at worst creates risk."

Only 3% of businesses consider themselves fully prepared for agentic AI, though 69% report satisfaction with current AI ROI. More than two-thirds remain unconvinced AI is achieving its full potential.

Kask emphasized that realizing AI value requires integrating systems with contextual data and processes while establishing robust governance—an approach SAP terms the "Autonomous Enterprise." The findings were first reported by SAP in its Value of AI Report 2026.