How widely you deploy AI doesn’t determine what it returns
Most of the Fortune 500 runs agentic AI, but only about a third of firms see it lift EBIT. Architecture and process fit make the difference.
Return on AI: Weekly Digest – September 9, 2026
Seven findings across manufacturing, retail, financial services and enterprise software, from faster product design and margin gains to cost-efficient model routing.
Our take
Eighty percent of the Fortune 500 runs agentic AI, but only 37% of firms see it lift EBIT. Deployment breadth isn’t the variable that matters. Globant shows what is: by routing requests intelligently across models, it grew usage 6.7 times while spend rose just 2.2 times.
The other results share a pattern. Westpac cut a six-week data migration to hours. Convenience stores saw a 2 to 5 percent margin lift from AI-assisted SKU planning. Midea compressed design cycles enough to reach new markets first. Each applied AI to a structured, high-volume workflow with a clear before and after.
What to do about it
- Measure AI programs by cost per output and EBIT contribution, not by adoption.
- Invest in routing and architecture so usage can grow faster than spend.
- Prioritize structured workflows where time-to-completion is easy to measure.