Connect OEE Data Before AI: Schneider Electric's Message
In an interview published by IIoT World on October 2, 2026, Schneider Electric's CPG lead argues that plants should connect their OEE data first and invest in AI only afterward. It is a vendor's view, but it touches real issues for anyone running a plant.
On October 2, 2026, IIoT World published a short article based on a video interview with Neil Smith, Segment President for Consumer Packaged Goods at Schneider Electric. The message is blunt: before talking about artificial intelligence, plants should fix their monitoring data. It is worth remembering that this is the view of a vendor executive, not independent research.
The problem: data that exists but stays siloed
According to the article, 18% of total product cost in the US CPG sector comes from delays and downtime. Yet many plants measure OEE (Overall Equipment Effectiveness) using paper, offline spreadsheets or disconnected software, and downtime reason codes are assigned manually by operators. The data is there, but locked in separate systems it cannot reveal the patterns behind recurring problems.
The data exists, but in isolation it does not reveal the causes of recurring problems.
The first priority: automate and connect
The first step described is to automate the assignment of reason codes and connect OEE systems to plant networks. Once connected, the data can be analyzed with data science techniques to correlate problems with other factors. The example offered is a recurring jam on a carton feeder, which might turn out to be linked to humidity on the floor, outdoor temperature or shift patterns.
Smith also describes advanced process control, long established in CPG plants, as the "first generation of AI", and argues that more sophisticated models rely on the same connected data infrastructure.
Closing the loop down to the machine
Understanding a cause is not enough; you need to be able to act on it. For plants with legacy control systems, the article notes, there is no native path from the enterprise AI platform to the equipment that carries out the change. This is where software-defined open automation (SDA) would come in as an "action broker": it could restart a line, adjust a setpoint or increase yield. Plants would keep their existing systems and migrate to a native SDA platform at their own pace. This is Schneider's commercial proposition, so it should be read as vendor positioning.
Operator knowledge
Smith points out that operators know the plant better than anyone. Building their experience into models could preserve that know-how as the workforce ages and hiring becomes harder. The end state described involves automatic pattern detection and cross-system causal analysis, with people supervising and approving rather than deciding every step.
Key takeaways
The article presents no case studies or ROI figures, and it does not cite a source for the 18% figure, so the numbers should be treated with caution. The order of priorities, however, is sensible for technical and IT leaders:
Practical questions to ask before starting an AI project:
- Are downtime events and their reason codes recorded automatically and consistently, or do they depend on manual entry?
- Is OEE data accessible on the plant network and comparable with environmental, process and shift variables?
- Is there a controlled path to feed a decision back into the control system, even on older equipment?
- Is operator knowledge captured in a structured form?
The theme is echoed by another piece linked from IIoT World, "CPG AI Adoption Outpaces Data Readiness" (September 25, 2026), which we have not read and therefore do not comment on. In short: monitoring and connectivity come first, AI second.