Why Enterprise AI Adoption Is Outpacing Measurable ROI

Enterprise AI’s central problem is no longer whether people will use it. It is whether organizations can connect what they spend to a measurable improvement in a business workflow. The 2026 surveys show substantial gaps in cost tracking and evaluation—not proof that AI cannot deliver value.232534

What the 2026 surveys actually show

Flexera reports that 31% of organizations have accurate visibility into AI software spending, while 59% say wasted AI spending increased over the past year.34 In Battery Ventures’ survey of 100 senior technology leaders, 94% said their organizations lacked a consistent enterprise-wide framework for evaluating AI ROI. Only 16% reported positive ROI on more than half of their AI projects.23

Budget performance tells a related story. In Futurum’s second-half 2026 survey, 46.9% of respondents said AI spending exceeded plan, compared with 5.6% who said it came in below plan.33 These surveys measure different groups and different things: visibility, perceived waste, project returns and budget variance should not be combined into a single AI failure rate.

Why adoption is easier to count than value

A license, an active user or a completed pilot is not a financial return. To establish one, a company needs a baseline for the work being changed, an agreed outcome—such as faster service or higher-quality output—and a way to compare the benefit with the full cost of running the system. KPMG found that, despite measures such as dashboards and approval processes, most organizations still lacked real-time, end-to-end visibility into AI-related operating costs.19 In its global survey, organizations with full operating-cost visibility were more likely to report established ROI than those without it, 15% versus 3%. That is an association, not proof that visibility alone causes returns.25

Measurement also depends on organizational change. Bank of America’s technology chief has identified end-to-end process transformation, reuse, governance and ROI as priorities for making AI projects work at scale.14 Meanwhile, a survey reported by CIO found that 61% of organizations had paused, delayed, scaled back or abandoned IT-modernization initiatives over the previous two years. Budget shifts toward AI were one reported pressure, but the figure does not establish that AI spending caused every delay.18

Agentic AI raises the stakes because costs, business value and risk controls must remain clear as projects expand. Gartner forecast in 2025 that more than 40% of agentic AI projects would be canceled by the end of 2027 for reasons including escalating costs, unclear value and inadequate risk controls. It is a forecast, not an observed 2026 cancellation rate.51

Bank of America shows both progress and the remaining gap

Bank of America says AI tools are available to about 200,000 employees and that coding productivity has improved by roughly 15%–20% among its approximately 19,000–20,000 developers.34 That workflow-specific result is more informative than an adoption count. The bank has also estimated approximately $800 million in benefits against $400 million in costs, but those are management estimates—not independent verification that each use case pays for itself.2 Its co-president has said the clearest gains so far are in technology work such as coding.1

The practical test is to give each proposed deployment an accountable owner, a before-and-after workflow measure, a complete cost ledger and a quality or risk check. Expand it when the measured benefit holds up after those costs—not simply when usage rises. That approach leaves room for genuine AI gains without mistaking activity for ROI.192325