Only 7% of Organizations Have Fully Scaled AI, McKinsey Survey Finds

The analysis, written by the chief executive and co-founder of the marketing data company Adverity, says that gap becomes clearest for marketing teams when a pilot meets the complexity of an existing data stack. A pilot can demonstrate what a model does when it has a defined task and carefully selected data; moving the same capability into a live enterprise environment is considerably harder.

The difficulty sits beneath the model itself. Marketing data is spread across platforms, warehouses and internal systems, each with its own structures and definitions, which makes it hard for AI to establish what the numbers actually mean.

Production adds complexity that pilots avoid. In an enterprise marketing stack, a model may encounter several versions of the same metric, carrying different field names and different rules for how they should be calculated. Without a governed understanding of that environment, the analysis says, the model can act like a black box and make its own assumptions.

Cost is given as an example. One platform may store it under one field name while another uses something completely different. To a person who understands the organization's data the distinction is obvious; to a model operating without that knowledge layer, the first plausible match can look like the right answer. The output that follows can still sound credible, leaving a business with a confident but incorrect insight until it reaches a decision.

Improving data quality is therefore only part of the challenge, according to the analysis. AI also has to understand the meaning attached to the data. For marketing teams that means establishing canonical definitions for important concepts, mapping those definitions across the platforms where they appear, and capturing the rules the organization uses to measure performance.

A knowledge layer between the data warehouse and the AI working with the data can encode how metrics should be interpreted, which fields correspond across platforms and which business rules need to be applied before an analysis is considered reliable. Such layers sit on top of existing warehouses, so organizations do not have to make costly investments in replacing their infrastructure.

Static definitions alone are not enough. Marketing performance changes alongside the business, and the analysis says AI also needs access to the context surrounding a particular investigation. A campaign may look as though performance has deteriorated when the underlying reason is a recent budget change, and a sudden movement in results may coincide with a new promotion or a live market test. Without that information, a system can identify a change accurately and still reach the wrong conclusion about what caused it.

The analysis suggests separating persistent marketing knowledge from context that is resolved for each investigation, and points to collaboration between marketing and data teams. Marketing leaders understand the objectives behind campaigns and how performance should be interpreted, while data leaders understand the systems, structures and controls through which that information is managed. Definitions can then be agreed before they become embedded in automated workflows, with governance kept inside the analytical process rather than introduced after an output has been generated.

The next stage of enterprise AI adoption, the analysis states, depends on making these foundations operational, and it cites McKinsey's finding that most organizations remain somewhere between experimentation and full scale.