The earliest period of enterprise AI adoption was characterized by unusually fast procurement. Budget approved for a pilot within weeks, minimal security review. Current research from McKinsey and Gartner shows why that era gave way to something closer to ordinary enterprise software procurement discipline, and the data cuts in genuinely two directions at once.
The ROI case is real, not hype
McKinsey's Global AI Survey 2025 found organizations reporting an average 5.8x ROI on AI investment within 14 months of production deployment, and companies using generative AI reported an average $3.70 return for every dollar spent. That's a genuinely strong headline number, and it's part of why enterprise AI spending hasn't slowed in aggregate even as individual procurement cycles have gotten longer and more scrutinized.
But scaling remains the harder problem than piloting
The same research shows nearly two-thirds of organizations haven't yet begun scaling AI across the enterprise, and only 39% report EBIT impact at the enterprise level despite widespread use-case-level deployment. That gap (strong ROI on individual deployments, but limited enterprise-wide financial impact) is the real story behind the procurement slowdown: it's not that AI doesn't work, it's that getting a successful pilot to scale across a large organization is a substantially harder problem than the pilot itself, and procurement teams have learned to price that difficulty into how fast they'll approve the next deal.
Gartner's failure-rate projection is shaping buyer caution directly
Gartner projects that 40% of agentic AI projects will fail by 2027, citing escalating costs, unclear business value, and inadequate risk controls as the primary causes, a widely cited figure that's become a genuine reference point in enterprise AI purchasing conversations, independent of whether any specific buyer's own pilot is at risk of that outcome. A number like that changes negotiating leverage and diligence depth even for vendors with a genuinely strong product, simply by resetting the baseline level of skepticism buyers bring into the room.
What buyers actually prioritize now, and it isn't price
When evaluating AI tools, enterprises now prioritize measurable value delivery (30%) and industry-specific customization (26%) far above price considerations, which sit at just 1% of stated priority. That's a genuinely sophisticated buying posture. It suggests procurement teams have moved past "is this affordable" to "can this vendor prove it will actually work for our specific context," which is a meaningfully higher bar for a vendor to clear than a compelling demo alone.
Where the friction concentrates
Regulated industries and large enterprises face procurement cycles, legal review, data residency requirements, and security governance that don't compress just because a technology is compelling, a discipline that's shown up directly in the enterprise legal AI adoption we've covered and the ambient healthcare documentation deployments that have scaled specifically because they were built to satisfy exactly this kind of review from the start.
What this means for AI vendors specifically
The practical implication: a compelling demo is necessary but no longer sufficient. Vendors who can walk into a sales conversation with a specific, defensible ROI case matched against McKinsey's and Gartner's now-widely-cited benchmarks, and clean answers to standard security and compliance diligence questions, are closing deals meaningfully faster than vendors still selling primarily on capability. That's a maturing market doing what maturing markets do (separating genuine, scalable value from early-cycle enthusiasm) and the data suggests it's a real, measurable shift, not just anecdotal caution.