Data Before AI: Schneider Electric's Advice to CPG Plants
According to Neil Smith of Schneider Electric, consumer packaged goods plants should first connect their OEE data and automate downtime reason codes. Only then does investing in AI make sense.
On 2 October 2026, IIoT World published an article by Lucian Fogoros based on a video interview with Neil Smith, Segment President for Consumer Packaged Goods (CPG) at Schneider Electric. The message is clear: before talking about artificial intelligence, plants need to fix the foundations of their data. The page notes that AI tools were used to summarize the content.
The problem: data that exists but says nothing
The figure cited is 18%: according to the article, that is how much production delays and downtime add to the cost of CPG products in the United States. However, the source gives no independent reference for this number, so it should be read as a stated order of magnitude, not a verified figure.
The picture described will be familiar to many readers: in numerous plants, OEE is still captured on paper sheets, offline spreadsheets or disconnected software. Downtime reason codes are assigned manually by operators. The data is there, but it sits in isolated systems and does not explain why the same problems keep coming back.
First priority: automate and connect
Smith names automatic assignment of reason codes and connecting OEE systems to plant networks as the first step. With connected data, data science can correlate, for example, a recurring jam at the carton infeed with humidity in the area, outside temperature or shift patterns. Without that foundation, such correlations stay invisible.
One interesting point is how he places AI in continuity with the past: Smith calls advanced process control "the first generation of AI." The more sophisticated models, he says, rest on the same connected data infrastructure.
Closing the loop: from recommendation to action
An AI recommendation, such as changing a setpoint or restarting a line, is only useful if it becomes action on the floor. According to the article, legacy control systems have no native path from enterprise AI to the machines. The proposal is to use open software-defined automation (SDA) as an "action broker" between the two worlds, running alongside existing systems while plants migrate to the native SDA platform at their own pace.
Operator knowledge
Smith argues that operators know the plant better than anyone. Capturing their experience in AI systems would help preserve know-how as the workforce ages and retires.
Why it matters for PLC, SCADA and MES integrators
For integrators the message is practical: the groundwork comes before AI. Specifically:
The activities the article puts first:
- connecting PLC and OEE data in an accessible architecture;
- automating downtime classification, reducing manual entry;
- building a secure path from AI recommendations to the controllers.
A note of caution
The article comes from a vendor source. The SDA recommendation comes from Schneider Electric, which sells it, and the 18% has no independent source. That does not make the underlying reasoning wrong, and many digitalization projects bear it out, but it suggests evaluating architectural alternatives critically.
The groundwork comes before AI: connected data, automatically classified downtime, a secure path to the controllers.
The practical advice, valid whichever vendor you use: check how many of your plant's stoppages have a reason code assigned automatically and how many depend on manual entry. That measure shows how ready you really are for AI.