Neil Smith of Schneider Electric and Andrew Waycott of TwinThread answer six audience questions on building the CFO business case for industrial AI, data readiness, OEM-operator data sharing, industry-specific adoption barriers, cybersecurity safeguards, and the organizational change required to move AI from a successful test to a trusted production capability. The questions were asked during the “From AI Pilots to Profit: Unlocking Manufacturing’s Hidden Data Advantage” session at the Industrial AI Summit 2026.
What Convinces a CFO to Fund AI: Cost Avoidance or Revenue?
Audience question:”When you build the business case for an AI project, what convinces a CFO more: the cost avoided (downtime, scrap, energy) or the new revenue enabled? And which one is easier to prove?”
Andrew Waycott, President and Co-founder of TwinThread:
I would say the initial focus is usually on cost avoidance. With the right analytics in place, CFOs already know exactly how much is lost to downtime, scrap, and inefficiencies; there is a concrete dollar amount attached to those problems. For example, with process manufacturers, TwinThread typically starts by reducing process variance. This is the primary driver of those losses. Over time, as we achieve more targeted optimization, we can push performance well beyond baseline expectations and ultimately redefine what target revenue should look like.
However, so much of it depends on the use case the manufacturer wants to tackle. And that’s subjective.
Neil Smith, Segment President, Consumer Packaged Goods at Schneider Electric:
It needs to be a balance of both. Our representative 2026 AI in CPG study reveals mean industrial AI ROI today is 24.7% in the U.S. (18.7% globally), expected to rise toward 50-74%+ by 2030 for the most mature AI deployments. However, only 13% of CPG manufacturers report AI fully embedded end-to-end today, so the revenue-upside case is still largely projected rather than realized. The key measure is whether your AI deployment is designed to be financially sustainable and operationally effective. Ultimately cost avoidance gets the project approved. Revenue growth gets the strategy funded.
How Much Data Do Manufacturers Actually Need to Start With AI?
Once the CFO approves funding, the next question is whether the plant’s data is ready for AI, or whether manufacturers need to solve a broader data accessibility problem first.
Audience question:”Should we look beyond Industrial AI because a lot of plants have not made their data accessible? Should we take a broader view with Industrial Intelligence?”
Andrew Waycott, President and Co-founder of TwinThread:
True value lies in solving specific, targeted manufacturing problems. Attempting broad, sweeping technology overhauls usually collides with hard-to-change realities: ingrained human behaviors, entrenched workflows, and rigid legacy systems. There is also a misconception about how much data is required to get started with Industrial AI. We’ve consistently proven that, on an extrusion line for example, just 10 to 30 parameters are often enough to significantly reduce startup scrap and deliver value quickly.
Neil Smith, Segment President, Consumer Packaged Goods at Schneider Electric:
Yes, absolutely. Industrial AI without industrial intelligence is simply sophisticated guesswork. The quality of AI outcomes will always be limited by the quality and context of industrial data. Our 2026 Industrial AI in CPG study found that 36.3% of manufacturers cite a lack of contextualized operational data as a barrier to adoption, so for many plants the first step isn’t AI at all. It’s making data accessible, connected and trustworthy. Get that foundation right, and AI becomes a natural next step rather than an expensive experiment.
How Do OEMs and Operators Share Data and Cut False Positives?
Starting with minimal data works for a pilot. Scaling across equipment, plants, and fleets requires continuous model tuning and data sharing between OEMs and operators, which introduces its own set of barriers.
Audience question:”Industrial AI pilots often show promise, but scaling them requires continuous model tuning using real operational data from equipment, unit, plant, and fleet levels. How have you successfully addressed data-sharing challenges between OEMs and operators, reduced false positives, and created commercial models where all parties share in the value generated?”
Neil Smith, Segment President, Consumer Packaged Goods at Schneider Electric:
It’s an important question, and it goes to the heart of why so many pilots stall. Most manufacturers aren’t short on data. Between existing automation, historians and quality systems, the information needed to run a smarter operation is usually already being generated on site. What’s missing is the connective layer that turns that scattered, siloed data, often locked into proprietary legacy systems, into a single, real-time picture. On data sharing, the biggest barrier between OEMs and operators is trust. Operators want the benefits of OEM expertise without losing control or ownership of their operational data. This is one of the clearest use cases for AVEVA CONNECT: it lets manufacturers share real-time data securely, inside and outside the enterprise. For example, they can give an OEM governed access to the specific equipment data it needs to analyze and optimize performance, while the operator keeps ownership and decides who sees what, and for how long. Open, software-defined automation strengthens this by decoupling control software from any single hardware vendor, so data is no longer trapped in proprietary systems. A recent study from Omdia found that modernizing closed, legacy systems could save a medium-sized business $11.28 million (£8.3 million) a year. These hidden costs, averaging 7.5% of annual revenue, stem from operational inefficiencies, downtime, compliance retrofits, and delayed production, which are issues often masked by the perceived reliability of legacy automation systems. Reducing false positives comes down to context. A model that sees a vibration spike in isolation will raise alarms. A model that also knows the product being run, the line speed and the maintenance history can tell a genuine fault from normal process variation. That’s why contextualized data from platforms like AVEVA™ PI System matters so much. Feeding operator feedback back into the model continuously is what turns a promising pilot into something teams rely on.
Is AI Hardest in Aerospace, Government, or Regulated Industries?
Data sharing and scaling barriers vary by industry. Some sectors face steeper regulatory and organizational hurdles than others.
Audience question:”Which industries are the most challenging when it comes to implementation of industrial AI and IIoT? Aerospace? Offshore oil exploration? Something else?”
Andrew Waycott, President and Co-founder of TwinThread:
Government organizations or any industry that goes through a lot of red tape to get products on the shelves (e.g., pharmaceutical companies).
What Safeguards Are Needed Before AI Agents Enter Production?
Regardless of the industry, expanding AI from a pilot into mission-critical operations raises cybersecurity and governance concerns that manufacturers need to address before scaling.
Audience question:”Many AI pilots demonstrate technical success, but the Hugging Face incident highlighted the risks that emerge when autonomous agents gain access to production environments. What safeguards should manufacturers establish before expanding AI pilots into mission-critical operations?”
Neil Smith, Segment President, Consumer Packaged Goods at Schneider Electric:
Manufacturers should ensure their Industrial AI delivers consistent, repeatable outcomes they understand and explain. I would reinforce that to scale data intensive operations, like AI, first a manufacturer needs to manage cyber risk. Cybersecurity is what makes digital transformation possible. Once your data becomes monetizable, it becomes valuable, and it needs to be managed accordingly. Cybersecurity is the trust layer for digital transformation. It gives organizations the confidence to adopt AI, cloud and automation, and to connect their systems at scale. The scale and speed of threats are outgrowing human capacity. AI can analyze large volumes of data, identify patterns and surface risk far faster than a human team can. But cybersecurity is about managing risk, and that requires judgment. The most effective model is AI-accelerated and human-led, where AI augments human expertise rather than replacing it. Look at detection and response. We moved from organizations not noticing incidents at all, to knowing about them but taking a long time to respond. What AI brings is speed of response, because it processes a much wider set of context, not just security data but operational data as well. AI becomes part of the autonomous operations stack, which means it is now an asset you have to manage. The risk around the AI itself, the data and the models belongs in the risk register alongside everything else. SE Advisory Services works with industrial organizations to harden OT assets, apply zero-trust principles and maintain continuously monitored environments, using an approach aligned with IEC 62443. Managed well, cybersecurity supports operational continuity and productivity rather than sitting on the balance sheet as a cost of operating in a regulated market.
What Organizational Change Turns AI Pilots Into Trusted Systems?
With the business case, data foundation, and safeguards addressed, the final barrier is organizational: what must leadership actually change to move AI from a successful test to a trusted production capability?
Audience question:”If we assume that the technology is already mature enough, what is the single biggest organizational change leadership needs to make to turn AI from a successful pilot into a trusted decision-making capability at scale?”
Andrew Waycott, President and Co-founder of TwinThread:
Operator adherence. A pilot can generate the perfect recommendations, but if personnel on the floor don’t execute them, zero value is realized. We encourage adoption through progressive levels of control: starting with human-in-the-loop, advancing to human-out-of-the-loop, and eventually reaching closed-loop automation with override capabilities. Over time, operators can confidently offload manual tasks to the system. By continuously embedding their feedback, the AI evolves into a true digital reflection of exactly how an expert would handle any situation.
Neil Smith, Segment President, Consumer Packaged Goods at Schneider Electric:
What we often see is that the scaling challenge is rarely technology but organizational adoption and execution. The biggest shift leaders need to make is treating AI as an operational transformation program, not a technology project, ensure senior leaders actively support your AI initiatives is key. Disciplined planning, a clear roadmap, and evidence-based validation before scaling are also a must. By testing use cases through pilots and applying digital twin-based modeling, we ensure every deployment is proven, repeatable, and scalable.
These questions were submitted by attendees of the “From AI Pilots to Profit: Unlocking Manufacturing’s Hidden Data Advantage” session at the Industrial AI Summit 2026. Answers provided by Neil Smith of Schneider Electric and Andrew Waycott of TwinThread. The session also featured François Coulloudon of TEEPTRAK Groupe, and it was moderated by Kevin O’Donovan. The session was sponsored by TEEPTRAK and TwinThread.
