Kostiantyn Gitko is the CEO of Devox Software, specializing in AI-native transformation and scaling complex systems for global enterprises.gettyIn the first two parts of this series, we covered the architecture: why resilience is now a competitive strategy, how edge intelligence distributes decision-making closer to the floor and how AI captures the expertise that walks out the door at retirement. Architecture, however, is only part of the equation. Metrics help reveal whether it actually exists. Once AI starts making calls on the floor, it's no longer a tool—it's operational infrastructure. And uptime and latency are just the beginning.Time: The Pace Of Recovery​Every AI system eventually reaches the limits of its training and encounters a situation it has never seen before. In my experience, that moment often comes sooner than the team expects. What happens next is what separates companies that automate from those that build real operational resilience.​I recommend measuring four things: detection, decision, action and recovery. This is where metrics like data drift matter, including how quickly data changes and how well model confidence matches actual accuracy. It also matters how consistently the system performs across every plant, because stability at one facility says little about the network as a whole.You've probably seen these metrics during prototyping, when teams like mine build demos that show how the system performs under real operating conditions. Worker safety relies on those milliseconds. In resilient operations, they separate a contained incident from a cascading failure.Data: Decision Advantage​More data rarely improves outcomes; better decisions do. AI shortens decision cycles, but without disciplined routing, you get signal overload and stalled execution. When every signal is treated as urgent, operators lose the ability to distinguish what matters. That's not intelligence—it's noise presented as insight.​According to McKinsey, "this bottleneck is a key reason only 7 percent of companies have fully scaled AI across their organizations." Fixing unstructured data alone will not solve the problem. AI readiness depends on connecting structured and unstructured data into a governed, reusable foundation. Otherwise, it's like putting a race engine on a gravel road.​A risk score of 0.71 means one thing to a data scientist and something entirely different to a plant manager deciding whether to stop a production line. Flagging a normal temperature of 82 as critical can send teams chasing problems that don't exist. The same is true for supplier data: A model can be technically correct yet operationally wrong if it's working from outdated specifications. The system needs to adapt as conditions change instead of waiting for the next rebuild. How quickly it adapts determines whether autonomy becomes an asset or a liability.Autonomy: Decision ProximityResolution rate, alert-to-action ratio and MTTR are easy to track. They tell you how fast the system moves, but they don't tell you when it should have stayed quiet.Manufacturers are already putting agentic AI to work, and based on my conversations, many report measurable operational gains. Automated visual anomaly detection has emerged as one early success, catching defects sooner on the production line. And the opportunity extends far beyond quality inspection. ​Agentic AI could "generate $450 billion to $650 billion in additional annual revenue" across advanced industries by 2030—a 5% to 10% increase. Manufacturing and automotive are expected to see some of the biggest gains. The technology could also reduce costs by 30% to 50% by automating repetitive tasks and streamlining operations.​Even so, AI still operates with incomplete context. ​Sometimes, on the factory floor, data drops or integrations fail. In those moments, a resilient system gives operators a unified view of every signal, helping them catch problems before they spread.Even when every system is online, another challenge remains. Some parts of the production process still happen outside digital systems, so not every decision is captured in data. Sure, end-to-end visibility helps, but it is only part of the answer. Operators notice what the data misses: a smell, an unusual sound or subtle changes in a machine's behavior. In practice, resilience systems pair AI speed with human judgment rather than forcing a choice between them.​The same principle applies to the AI stack. The real lifecycle question is whether today's solution can adapt to tomorrow's. Can you safely replace, roll back or retire it when a better option becomes available? Given how quickly the industry is moving, that day will arrive sooner than most deployment teams expect.Economics: Resilience As MarginThe economics of resilience have changed. What was once an insurance policy is now a performance lever. The business case comes down to two numbers: the cost of failure and the cost of prevention. Most manufacturers only know one of them.Traditional preventive maintenance followed the calendar instead of machine condition. Healthy assets were serviced too early, while failing ones slipped through the cracks. The convergence of IIoT and AI changed that equation. Predictive maintenance tells you what is likely to fail. Prescriptive maintenance tells you what to do next.Energy efficiency tells a similar story. As AI workloads expand, the cost of running intelligence is becoming part of the manufacturing equation. Google, for example, reported substantial improvements in the efficiency of Gemini prompts during 2025, yet overall data center electricity demand continues to rise because AI usage is growing even faster. The International Energy Agency projects that global data center electricity consumption will more than double by 2030. For manufacturers, resilience increasingly depends not only on reducing downtime but also on deploying AI efficiently enough that its operational benefits outweigh its energy costs​​.Today, the metrics that matter are business outcomes: fewer defects, less downtime and lower energy consumption per unit. For manufacturers who get the metrics right, resilience becomes the margin.ConclusionManufacturing has reached an inflection point. McKinsey’s research on scaling AI in manufacturing shows that organizations successfully deploying AI at scale are realizing productivity improvements on the shop floor, directly supporting higher performance rates within OEE.Successful organizations are building capabilities that compound in value over time. I believe the advantage belongs to manufacturers that use AI to sharpen human judgment, recover faster and come back stronger after every disruption. That's antifragility, and it shows up on the bottom line.​​​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?