The models got demonstrably smarter this year. Enterprise ROI did not move. That's the finding buried in McKinsey's latest global survey of 1,719 leaders across 97 countries, and it should kill the most persistent myth in enterprise AI: that capability solves the value problem.
The myth: once AI is good enough, ROI follows
Scaling is up: 44% of organizations now deploy AI at enterprise scale, up from 38% a year ago. Employees feel it — 80% report individual productivity gains. But only 37% of leaders can attribute any positive EBIT effect to AI, and just 6% qualify as "high performers." Both numbers are flat year over year. Deloitte China found the same shape: production deployments rose to 64.4%, yet 48.5% of firms report returns below expectations and 44.4% can't calculate ROI at all.
The bottleneck moved from answering to operating
Smarter models reason better. They don't reconcile duplicate customer IDs across your CRM and ERP. They don't handle exceptions, enforce permissions, or write back auditable records when an API fails halfway through a job. That's why 73% of high performers redesign workflows around AI, versus 25% of everyone else. The gap isn't model quality — it's whether the saved hour gets re-deployed into higher-value work or simply absorbed.
Measure the process, not the tool
Deloitte's data shows why the numbers stay murky: 60.4% of enterprises have no value-measurement framework, and 87.3% grade AI on processing speed and hours saved. That's activity, not outcome — prompts and active users instead of days-sales-outstanding or cost per transaction.
The takeaway: stop waiting for a better model to fix your P&L. The next advantage goes to leaders who rewire workflows and measure the business outcome — not the benchmark score.