Governance and cost control now decide which AI projects reach production
Most stalled AI programs aren’t failing on technology. They’re failing on governance gaps, surprise costs and experts who drop out before launch.
Return on AI: Weekly Digest – August 11, 2026
Twelve findings across governance, finance, workforce readiness and research operations, including rising token costs, cancelled projects and AI-native finance workflows.
Our take
This week’s stories describe AI programs stalling for reasons that have little to do with the technology. Governance gaps delayed almost every firm Cloudera surveyed. A quarter of firms cancelled projects after token costs arrived higher than planned. And the most common reason pilots never reach production is that the domain experts who shaped them move on.
The contrast is OpenAI’s own finance team, which is redesigning its close process around AI rather than adding AI to the old one. That kind of redesign needs governance and cost controls in place from the start.
What to do about it
- Set usage budgets and model routing before you scale, not after the first invoice.
- Keep the domain experts on each project involved through launch, not just discovery.
- Treat governance as a prerequisite for production, not a review at the end.