Manufacturers are under more pressure than ever. Workforce shortages, shrinking margins, rising quality demands, and unrelenting customer expectations are converging, forcing the industry to rethink how it operates. The good news? Artificial intelligence (AI) is emerging as one of the fastest, most effective ways to tackle these challenges head-on.
AI is already delivering significant value across the shop floor. Companies are using it to optimize production, strengthen quality management, reduce delays, minimize downtime, and capture institutional knowledge from experts nearing retirement. In a sector where minutes matter and mistakes are expensive, AI’s ability to boost efficiency, productivity, and profitability is undeniably transformative.
But there is a catch. And it is a big one.
Recent academic and industry research from MIT and others suggests that the vast majority of generative AI initiatives fail to progress beyond pilots or fail to deliver sustained business value. With billions already invested, that failure rate is unsustainable for any industry, especially manufacturing, where profitability is closely linked to eliminating inefficiencies and operational disruptions. And for complex discrete manufacturers, such as the Aerospace and Defense (A&D) sector (which my company serves), the cost of a failed AI program can reach hundreds of millions of dollars.
For our industry, AI failure isn’t an option. So why are so many of those AI initiatives struggling? It comes down to two intertwined mistakes: 1) Using AI that isn’t purpose-built for manufacturing in their industry, and 2) Trying to run AI on data that isn’t truly ready for it. Correct these two issues, and AI becomes a competitive advantage, fast.
Why Manufacturers Need Purpose-Built AI
Most general-purpose AI models and tools are not designed to handle the execution-level complexity of industrial manufacturing operations, where sequencing, traceability, and cross-system coordination are critical. Large Language Models (LLMs) trained primarily on consumer internet data may perform well for general conversational tasks, but they struggle with the complexity of industrial manufacturing environments—such as complex bills of materials, time-sensitive workflows, engineering-grade traceability and cross-system coordination.
To succeed, manufacturers need purpose-built AI with models trained on industry-specific data and engineered for sector-specific workflows. Many experts now refer to this approach as vertical AI. And for good reason: Discrete manufacturing is fundamentally different from continuous manufacturing, which is different from aerospace and defense.
On the surface, these tools share similar features, including chatbots and reporting features that teams are already comfortable with. But under the hood, a purpose-built AI for A&D, for example, doesn’t just provide these features, it is trained on sector- and company-specific engineering and manufacturing data, architected around PLM, ERP, and MES processes, and tuned to the rhythms of real production environments.
The specificity of purpose-built AI mitigates the failures identified by the MIT study. But purpose-built AI does much more than avoid failure. When AI is embedded across PLM, ERP, and a purpose-built MES, the digital thread strengthens—connecting design intent, execution context, and quality outcomes in real time. This enables AI to:
- Identify defects earlier in the manufacturing lifecycle and resolve issues before they impact delivery timelines
- Recommend solutions based on real-time and contextual data
- Drive consistent decision-making across plants and programs, ensuring consistency and confidence
- Automate repetitive tasks such as discrepancy triage, inspection record review, and work-instruction validation to increase productivity and reduce human error.
- Enforce global standardization across an expanding production footprint
This is the kind of performance that generic AI cannot deliver because it was never designed to.
Why AI-Ready Data is the Second Critical Ingredient
Even the best AI models fail when the underlying data isn’t ready. MIT’s Sloan School put it bluntly: “The lack of universal industrial data has been [a] major obstacle slowing the adoption of AI among mainstream manufacturers.”
Across the industry, too much data remains trapped in paper binders, siloed systems, or static repositories—and critically, lacks execution context such as who performed a task, under what conditions, and why decisions were made. Without clean, connected, high-quality, real-time data, AI will struggle, or worse yet, mislead.
To fix this, manufacturers must build a foundation of AI-ready data through three key steps:
- Deploy IoT sensors across shop-floor machinery to capture accurate, real-time production signals that reflect actual operating conditions. This enables the capture of accurate, real-time, dynamic production data, which will fuel your AI initiatives.
- Feed that data into a Manufacturing Execution System (MES) that contextualizes raw signals into structured execution records tied to specific operations, parts, and personnel. MES transforms raw machine data into structured, contextualized operational intelligence. And when the MES is purpose-built for your type of manufacturing, it enables AI analysis to drive more valuable insights.
- Integrate MES with ERP and PLM. This creates the rich digital thread needed for high-performance AI. This connection allows the system to contextualize work instructions, trace back to design intent, structure data for every execution step, and document variation and quality signals.
With this infrastructure in place, AI becomes not only feasible but also powerful.
Building the Digital Thread that Unlocks AI’s Full Potential
Manufacturers that combine purpose-built AI with AI-ready data unlock capabilities that were previously out of reach. These include:
- Adaptive manufacturing that increases efficiency, lowers costs, improves safety, and more
- Predictive quality that can prevent quality issues before production and that can achieve significant progress toward zero-defect manufacturing
- Self-improving processes that increase quality and efficiency
- A seamless digital thread that flows from design to production
- Workforce augmentation that helps new employees perform like highly experienced operators
For manufacturing, this data strategy isn’t just about avoiding the 95% failure rate. It’s about future-proofing manufacturing for the next decade of competitiveness.
The Bottom Line
AI will be one of the biggest drivers of manufacturing performance in the coming years, but only for companies that implement it effectively.
The winning formula is clear.
Purpose-built AI + AI-ready data = Sustainable, scalable AI success.
Manufacturers who invest first in purpose-built systems and execution-grade data won’t just keep up—they’ll be positioned to lead as AI reshapes industrial performance.
About the Author:
Sung Kim is the Chief Technology Officer of iBase-t. He is a computer scientist, product architect, and educator with over 20 years of experience in A&D and high-tech manufacturing. He leads iBase-t’s product and technology strategy, focusing on scalable, integrated solutions for complex manufacturing. He earned his B.A., M.A., and Ph.D. from the University of Texas, and he has held multiple academic positions at universities, teaching the next generation of leaders in the technology and manufacturing industries.
