Factory AI: The Real Bottleneck Is Data Governance, Not Models

On September 14, 2026, Engineering.com published an article on the manufacturing findings of Cloudera's 2026 Data Readiness Index. Cloudera is a company that offers data and AI platforms, and it had released the report itself on September 8. The underlying message: manufacturers still struggle to take AI from the data layer to the production floor.

What the numbers say

The figures point to a gap between knowing about data and controlling it:

• 82% of manufacturing respondents know where their data resides;
• only 58% say that all or almost all of their data is fully governed;
• 20% of organizations cite weak integration of AI and analytics into operational workflows as the main reason for missing the expected return on investment.

These are vendor-reported figures, and the material available does not allow us to determine the sample size or the survey methodology. They should therefore be read as an indicator of a trend, not as a definitive measurement.

The context: data everywhere, but fragmented

Manufacturers handle growing volumes of operational, production, supply chain, customer and IoT data. This data is spread across plants, supply chains, enterprise applications and edge environments, and that dispersion makes it harder to use consistently. Knowing where a piece of data lives is a first step, but it is not the same as being able to trust it: you need to know who owns it, how reliable it is, how it is defined and who can access it.

Knowing where the data is is not enough: you must be able to govern it and turn it into operational decisions.

From insight to decision

According to the summary of the article, the hard part is turning AI-generated insights into operational decisions. Integrating, governing, unifying and operationalizing data remains an ongoing challenge. A model that predicts a failure or suggests a process parameter delivers value only if the result reaches whoever, or whatever, has to act on it: an operator, a supervisory system, a production schedule.

Why it matters for automation professionals

For those working with PLC, SCADA and MES, this finding fits a practical reading: the bottleneck does not appear to be model quality, but data governance and the link between AI and the systems that actually run the process. In practice, this means addressing aspects that are often barely visible:

• tags and signals with consistent names, units and meanings across different plants;
• traceability of the origin and quality of the data feeding the models;
• clear interfaces between the control layer, supervision, MES and analytics platforms;
• rules on who can write a value or setpoint suggested by AI, and with what validation;
• defined responsibilities for the data lifecycle.

A practical takeaway

Before launching a new AI project, it is worth verifying the complete path of the result: which data it comes from, who governs it, which operational system it reaches and who decides to act. Defining this flow at the outset, together with return metrics tied to concrete decisions, reduces the risk of ending up with an accurate model that is isolated from the process. This is where industrial software skills and integration expertise across automation layers can make the difference.

What we don't know yet

The text of the article available to us was truncated, so the second main challenge it highlights and Cloudera's recommendations are unknown. It is advisable to consult the full report before drawing more detailed conclusions.