Most AI PoCs don't fail in the lab. They fail in the operating model.

Only 6% of organizations capture significant value from AI (McKinsey, State of AI 2025, n=1,993).

Adoption is universal. Value is rare.

88% of organizations use AI in at least one business function.
Two thirds have not yet begun scaling AI across the enterprise.
6% qualify as high performers, attributing more than 5% of EBIT to AI.

88%

use AI in at least one function

McKinsey · 2025

~6%

high performers (≥5% of EBIT from AI)

McKinsey · 2025

95%

of GenAI pilots show no P&L impact

MIT NANDA · 2025

42%

abandoned most AI initiatives in 2025

S&P Global · 2025

Figure 1. Four numbers from 2025: massive adoption, scarce value.

In the lab vs. the real enterprise

In the lab	In the enterprise
Data is available	Pipelines aren't stable
Assumptions are controlled	Processes are fragmented
The model performs	Ownership is unclear
The demo convinces	Security arrives too late
The sponsor sees potential	MLOps is immature
The team celebrates	No one is accountable

The real failure point

AI initiatives don't lose momentum in the algorithm. They lose it in the operating model.

This is most visible in asset-intensive industries: mining, manufacturing, energy, healthcare, food production. AI doesn't create value living in a notebook or a dashboard.

Eight connectors from model to enterprise value

Business priorities
Operational workflows
Decision rights
Data ownership
Model monitoring
Risk controls
Change management
Financial impact

Figure 2. Eight connectors between a model that works in the lab and value in the enterprise.

The right question

Stop asking: "Can we build the model?"

Start asking: "Can we operate, govern, scale and improve this model in production?"

AI value isn't captured in the lab. It's captured when intelligence becomes part of the operating system of the company.

What's the #1 reason AI initiatives stall in your organization?