More than half of large enterprises are stuck in an AI spending trap where their investments aren't generating measurable returns, according to new research from Domino Data Lab. The finding is particularly striking because it hasn't budged since 2025, even as 93% of companies report improved AI production capabilities. The disconnect reveals a hidden crisis in enterprise AI: companies can build and deploy AI models, but they struggle to translate those models into business value.

What's the Real Problem With Enterprise AI ROI?

The issue isn't that AI models aren't working. In fact, the research surveyed 639 senior AI leaders across North America, the United Kingdom, and continental Europe and found that enterprise AI production is climbing steadily. The problem is what researchers call the "last-mile gap." A model sitting in production is useless if the business users who need its insights can't actually access them.

When asked how business users actually access AI-generated insights today, the responses paint a fragmented picture. The most common scenario, reported by 34% of organizations, is a chaotic mix of different access methods depending on which business unit you work in. Even worse, 40% of companies still rely entirely on mediated access, meaning business users have to submit requests to data scientists or analysts who run the analysis and return results manually.

"Getting a model into production used to be the milestone that mattered. Our research shows that's not enough anymore. The real milestone is the moment a business user can act on what the model found, and for too many enterprises, that moment still isn't happening at the pace or scale of business," said Thomas Robinson, Chief Operating Officer at Domino Data Lab.

Thomas Robinson, Chief Operating Officer at Domino Data Lab

This last-mile gap explains why ROI remains flat even as production capability improves. Companies are investing heavily in AI infrastructure and data science teams, but they're not building the applications and interfaces that would let business users actually use those models to make decisions faster or better.

Where Is Agentic AI Heading, and Why Does Governance Matter?

Enterprise leaders are increasingly excited about agentic AI, which refers to AI systems that can take actions autonomously rather than just providing recommendations. Expanding agentic AI use ranks as the top organizational priority for 2026, tied with upskilling business users, both at 38.5%.

But here's the problem: many companies are deploying agentic AI without the governance infrastructure to manage it safely. The research found that 43% of organizations have agentic AI running in governed production, while 41% are piloting or scaling agentic AI without governance in place. Among those scaling without governance, the number is more than double those merely piloting.

Governance maturity is the clearest dividing line between success and failure. Organizations with fully integrated AI governance are 3.9 times more likely to have agentic AI running in governed production compared to organizations where governance is only partially keeping pace. Among companies with fully integrated governance, 75% report significantly improved AI delivery velocity, compared to just 23% of organizations where governance is falling behind.

How to Build AI Governance That Actually Works

• Build governance first: Financial services, banking, and insurance organizations lead every vertical on both governance maturity and production velocity because they built governance infrastructure before scaling AI, rather than retrofitting it afterward.
• Treat agents as managed entities: Agentic AI systems need to be auditable and controllable throughout their lifecycle, not just at deployment. Organizations that treat agents as managed, governed systems from the start are far more likely to scale them safely.
• Create direct access for business users: Rather than relying on mediated access through data scientists, successful organizations build governed AI-powered applications that give business users purpose-built interfaces to act on AI insights directly.

How Does This Problem Vary Across Regions?

The ROI plateau and governance gaps aren't uniform across geographies. North American organizations are markedly less likely to report ROI stuck at the same level as investment or lower, at 51.1%, compared to 66.9% in the United Kingdom and 67.0% in continental Europe.

However, North American organizations face a different problem: they're far more likely to report that business users have no direct access to AI-generated insights at all. Only 12.8% of North American organizations report no direct access, compared to 1.4% in the UK and 6.4% in Europe. This suggests that North American companies may be more likely to have built some kind of access layer, even if it's not optimal.

The agentic governance gap is most acute in Europe. European organizations report the lowest rate of fully integrated governance at 42.6%, compared to roughly 51% in North America and the UK. That shortfall shows up directly in agentic deployment: nearly half of European organizations are piloting or scaling agentic AI without governance, compared to 40% in North America and 38% in the UK.

The broader implication is clear: the companies winning with AI aren't necessarily the ones with the most advanced models or the biggest budgets. They're the ones that solved the last-mile problem by building governance early and creating applications that let business users actually use AI insights to make decisions. For the 57% of enterprises still struggling with ROI, the path forward isn't better models. It's better delivery.