Humatica study finds AI operating models lag enterprise adoption

A new Humatica whitepaper produced with CoreWeave finds that large enterprises are increasingly dependent on AI, while formal structures for governing and owning those systems remain limited. The research identifies the organization, rather than technology, as the main constraint on progress.
- 95% of large enterprises have moved beyond AI experimentation.
- Only about a third have a formal operating model for AI.
- 68% of leaders cite organization and governance as the main constraint.
- One in five enterprises has named an AI leader.
- Humatica identifies five practices for creating lasting AI value.
AI is becoming critical to enterprise operations
The whitepaper draws on more than 30 interviews with senior executives and investors, alongside a panel of 252 enterprises across 14 industries. According to the research, two in three large enterprises already depend on AI, but only about a third have a formal AI operating model.
Humatica found that 95% of large enterprises have moved beyond experimentation. 66% depend on AI for a process, product, or business model, while 44% use AI in products they sell or in the economics of their business. For 12%, the business could not deliver its current offer at its current price or cost base without AI.
The report describes the gap between dependence and formal organization as the main AI risk. It also says 68% of leaders identify organization and governance as the main constraint to progress, compared with 28% who cite technology and infrastructure.
Ownership becomes harder as AI tools spread
One in five enterprises has named an AI leader, according to the study. These leaders help decide what gets built, who owns each system after launch and scaling, and who pays for it. The research says the objective is sustained competitiveness and profitability, rather than a short-term efficiency gain.
AI development is also becoming more decentralized. Of the AI use cases in production covered by the report, 42% started in the business, and one in four began with an individual or team acting on its own initiative. That makes it easier for employees to create tools, but raises questions about security, monitoring, evaluation, access, cost and continued improvement.
Participants told the researchers that building had become easier while owning had not. The report says leaders are responding with shared platforms that include controls for security, access, monitoring, evaluation and cost, allowing new applications to inherit those capabilities.
Five practices for building lasting AI value
Humatica points to five practices for enterprises making AI a core part of their organizations: treat AI as an operating model decision, enable decentralized building while scaling proven results, turn enterprise knowledge into a proprietary AI advantage, retain control of business-critical AI, and ensure every production system is owned, measured and improved.
Among AI leaders, 73% cite IP protection and differentiation as a reason to move a critical workload onto infrastructure they control. 36% cite cost and predictability. The report says these priorities become more important as companies use proprietary data, expertise and other intellectual property to create products and capabilities.
CoreWeave links the findings to its AI infrastructure products, including CoreWeave Forge, CoreWeave Inference, the CoreWeave Zero Trust AI Framework and CoreWeave ARENA. The company says Forge supports the continuous AI loop of running, observing, curating, improving and evaluating models or agents. The findings suggest that enterprises seeking lasting value will need clear leadership and governance alongside the technology used to build and operate AI systems.
CoreWeave: How an AI Operating Model Turns Ambition into Lasting Advantage


