The Human Side of Industrial AI: Accountability, Operators and Change

On October 15, 2026, IIoT World published answers to six audience questions gathered during the session "From AI Pilots to Profit: Unlocking Manufacturing's Hidden Data Advantage" at the Industrial AI Summit 2026. The respondents were Neil Smith, Segment President Consumer Packaged Goods at Schneider Electric, and Andrew Waycott, President and co-founder of TwinThread. The topic is one that often decides the fate of projects: people.

A necessary caveat: both speakers are technology vendors, so their views are not independent, and the piece is a panel Q&A, not research. It remains useful as a summary of practices that many technical leaders are grappling with.

Workforce resistance is a measurable risk

According to Schneider's 2026 study on AI in the CPG sector, 25.7% of manufacturers worldwide cite workforce resistance as a risk to scaling AI. It is not a marginal issue, but an item that belongs in the project risk analysis.

Recommendations to the operator, not autonomous actions

Waycott describes TwinThread's approach: the system detects process drift and sends the operator a recommendation instead of acting on the process by itself. The operator can approve, modify or reject it, and the model learns from that choice. Operators' historical actions define the initial setpoints and operating limits, while engineers use the feedback to improve the models.

The benefit is twofold: control stays with the people who know the plant, and operators' expertise becomes part of the improvement loop rather than being bypassed.

Who is to blame when the AI gets it wrong?

Asked who is responsible for a failure when AI acts without human supervision (the data engineer, the model vendor or the operator), Waycott answers that it is a failure of the workflow, not of a single actor. He proposes shared safeguards:

• data engineers ensure a clean, reliable data flow;
• model vendors define thresholds that alert the team when a failure is imminent;
• engineers and operators have clear procedures to spot and diagnose anomalies and to intervene.

Change management as part of the project

For Smith, industrial AI is not just an IT project. His practical advice is to involve operations and maintenance teams early, starting with the choice of use cases, and keep them involved through to release. Successful programs, he argues, invest in change management and skills as much as in platforms, and companies with the best return on AI tend to show higher employee engagement and retention.

Both also see AI as a tool for capturing veterans' experience and making it available to younger, more digital-native colleagues, easing the effects of turnover and shift changes.

The Wuhan example, with caution

Schneider cites the "Lighthouse" factory in Wuhan, where agentic AI tracked skill gaps and assigned personalized training. The figures are reported by the company and have not been independently verified:

• workforce readiness from 20% to 76%;
• 56% of employees reskilled;
• new product introduction time cut from 36 to 12 months, a reduction of 66.7%.

What to do in practice

For engineering and IT leaders, there are three takeaways:

• design AI as decision support, with operator approval;
• define shared accountability across data, model and operations, with written thresholds and procedures;
• budget for change management alongside the technology.

These are common-sense principles, but the gap between stating them and applying them is where many pilots stall. Treating operators as co-designers rather than end recipients is the most concrete way to reduce that risk.