For three years, corporate AI spending lived inside research and innovation budgets — discretionary, flexible, and sheltered from the rigour applied to operational expenditure. That has changed. A Q2 2026 survey by Gartner of 240 CFOs at companies with revenues above $500 million found that 67 percent have moved primary AI spending into operational budget lines, with formal ROI requirements attached.
The shift has practical consequences for how enterprises buy and deploy AI. When spending is discretionary, procurement decisions are made by technology teams with loose accountability. When spending is operational, finance teams require unit economics: cost per transaction, quality improvement metrics, headcount offset ratios. Vendors selling to enterprise AI buyers report that sales cycles now routinely include CFO-level scrutiny that was absent in 2023 and 2024.
The metrics enterprises are applying vary by function. In customer service, the primary measure is cost per resolved ticket compared with human agent handling. In software development, it is pull requests merged per developer per week, a proxy for productivity. In document processing and legal review, it is hours saved per workflow type. Marketing functions are measuring content production cost per approved piece.
The difficult finding from the Gartner survey: 44 percent of CFOs report that AI initiatives in their organisations have not yet hit the ROI targets set at deployment time. The gap is most common in two categories. First, projects where the primary benefit was described as "productivity improvement" without a specific plan for redirecting the freed capacity — the productivity gain exists but does not reduce headcount or increase revenue. Second, projects involving unstructured data at scale, where the accuracy requirements for business processes are higher than initially modelled, requiring more human review than anticipated.
The implication for AI vendors: the procurement conversation has permanently shifted from capability demonstration to economic proof. Pilots that cannot show a measurable unit cost improvement within 90 days are being terminated rather than renewed. The market is becoming more demanding, which will accelerate consolidation around the vendors who can deliver quantifiable outcomes over those who sell on capability benchmarks.