Beyond the Hype – Where AI Actually Works in Manufacturing

AI is being positioned as the next industrial revolution, but many leaders are struggling to translate enthusiasm into practical outcomes. More than 80% of U.S. manufacturing leaders say they plan to increase their use of AI within the next two years, but that isn’t always being done with a clear strategy for what they expect to realistically achieve.

Between board-level pressure to do something with AI and the reality of complex production environments, there’s a growing gap between expectation and execution. That gap needs to be closed if manufacturers hope to see the benefits of AI in their businesses.

AI adoption is outpacing readiness

Manufacturing leaders today know they need to adopt AI to stay competitive. A recent survey from The National Association of Manufacturers (NAM) found that 51% of manufacturers are currently using AI and that 80% will consider AI essential to expanding or maintaining their business.

Manufacturers are using this technology in all sorts of different ways. AI has massive potential. However, in many organizations, investments in AI aren’t delivering the promised ROI.

A recent survey by PWC and the Manufacturing Institute found that the push from leaders to adopt AI can outpace readiness for AI-driven change. That same survey also found, in their words:

• Uneven AI adoption can slow integration into daily operations
• Advancing along the AI adoption curve requires training and experiential capability building
• Human and readiness gaps can constrain AI adoption
• Limited frontline leader input can constrain AI adoption and execution

Finding the fix

Rather than treating AI as a universal solution, leaders must distinguish between applications where AI excels – such as improving human-system interaction, capturing operational knowledge and simplifying complex workflows – and areas where it remains poorly suited, including precision control and repeatable decision-making in production environments.

A great place to apply this is during your sales process.  Many manufacturing companies want to improve the guided selling process using AI. They should start by focusing on the customer’s goals or outcomes rather than asking them to choose specific parameters about a product.

Here’s a personal example: I recently bought my first electric vehicle (EV). I went to an EV manufacturer’s website, and the first thing I saw were the technical parameters, such as Range: 360-558 km. That’s a big range, but I understand what it means. But horsepower? No idea. And consumption? Now I really didn’t understand that; is consumption good or bad? Will it make a difference to my family’s annual ski trip, which currently takes me around seven hours? That’s the kind of information that would actually help me decide.

Having an AI to discuss my care-abouts with would help the EV maker understand what’s important to me and translate those care-abouts into the technical specs that are right for me.

Another example is to use AI agents to break complex problems into smaller pieces. Agents plan and execute multi-step goals. To get the most out of these agents, the tasks should be language- heavy.

Some engineering organizations are taking this approach when it comes to managing complex requirements. Sometimes the information is spread across hundreds or thousands of documents, and the format isn’t standardized. If you were to build a traditional software system to import all this information, there would be far too many rules and exceptions to those rules. Since AI is great at pattern recognition, you don’t need to write down every single rule about how to interpret those documents.

Best practices for AI in manufacturing

Break large problems into smaller tasks. AI has limited working memory. If you give it too much data or an overly broad problem, it can lose context and begin hallucinating as it attempts to still provide an answer. To reduce this risk, divide large tasks into smaller components and aggregate the results — an area where agentic frameworks are becoming especially effective.

Use AI for pattern recognition, not authoritative reasoning. AI is good at pattern recognition and identifying relationships, but it’s not good for logic and reasoning. For instance, their “reasoning” and math abilities are prediction, not computation. So, outputs can appear insightful, but they’re not guaranteed to be correct, unless you connect to another system that can validate the results. In this approach, the AI can orchestrate the workflow, but you’ve made an explicit decision about where the operational boundaries are.

AI in manufacturing can create real value, but there are specific scenarios where the risk outweighs the benefit. Those include safety- and quality-critical operations. It also includes low-data or highly novel scenarios because AI performs poorly when it hasn’t seen enough relevant data. Highly regulation environments are another challenge because AI still lacks true traceability. Real-time control systems with tight tolerances are another place where AI should be avoided given that AI latency or instability can be unacceptable in high-speed processes

From hype to practicality

AI isn’t a panacea for all manufacturing ills. But by considering the real-world use cases and common misconceptions discussed above, manufacturers can adopt a more balanced approach to AI. In the short term, the key point is to use AI for what it’s best at: problems that involve a high degree of pattern recognition. Don’t use it for cases that require predictability, consistency, logic or calculations.

Manufacturers can cut through the noise and identify where AI genuinely delivers value today. The goal is not to slow adoption but to ensure organizations deploy AI where it meaningfully improves productivity and decision-making rather than simply following industry hype.