The scaling gap

79% adopt, 11% reach production; 78% pilot, 14% scale. Five gaps account for 89% of the failures, and none of them are model capability problems.

The three diagnoses behind the funnel: the misframing, the horizontal/vertical imbalance, and the data foundation that only 5% believe is ready.

The funnel

The reasons pilots do not become production have by now been documented fairly substantially. 79% of organizations report adopting AI agents but only 11% are in production. 78% have agent pilots but only 14% have reached production scale.

Five gaps have been identified as accounting for 89% of scaling failures in one analysis: integration complexity, inconsistent quality at volume, absence of monitoring, unclear ownership and insufficient domain data. None of these are model capability problems. They are all organizational and operational.

The misframing

The convergent finding across multiple sources is that the majority fail to move beyond pilots because they treat transformation as a technology project rather than organizational redesign. Teams that do succeed invest equally in people, process and platform. As such we must conclude that success is defined less by how much AI is deployed and more by how well governed agentic AI is embedded into core workflows.

The imbalance

McKinsey adds a structural diagnostic to the discussion by defining a horizontal/vertical imbalance. Copilots and chatbots scale easily but deliver diffuse gains. Vertical use cases with domain-specific agent deployments are transformative, but fewer than 10% make it past pilot. Vertical use cases face six barriers to scaling: siloed teams, data gaps in vertical domains, cultural resistance, LLM limitations for domain tasks, lack of packaged vertical solutions and insufficient CEO sponsorship.

The data foundation

The last critical issue in scaling is the data foundation. Only 5% of enterprises say their data is ready to support AI at production scale, despite near-universal investment. This is expressed vividly in five core failure points: fragmented data ecosystems, lack of business ownership, pilot-driven culture, poorly designed HITL architectures and insufficient governance. Organizations are building sophisticated agents on foundations that cannot support enterprise deployment.

People, governance, technology architecture and data must all scale together in order for agentic AI to be successful. Organizations are advised to follow the dual transformation path, continuing to extract gains from horizontal use cases while investing in vertical ones.