Forging Trust: Lessons on AI Adoption from the Iron & Steel Industry
We share insights from the iron and steel industry on why AI adoption is lagging and how developers can bridge the gaps to make AI more usable in traditional sectors.
Despite years of hype, AI adoption in traditional sectors like manufacturing has fallen short of expectations. Workers aren’t using it – in 2024, just 22% of manufacturers in the materials sector reported using AI (McKinsey). And employers aren’t demanding it – only 3.8% of job postings in the manufacturing industry required AI skills (Artificial Intelligence Index Report 2025). The heralded AI productivity revolution remains largely undetectable, especially in the field of generative AI (The Economist).
Nowhere has this been more apparent than the iron and steel industry. At the 2025 Association for Iron & Steel Technology Conference (AISTech), Roundtable interviewed over 50 engineers, operators, and executives. The conversations on the ground painted a picture even starker than the stats: fewer than 10% reported using generative AI for anything beyond editing emails. The adoption of AI has been “cold rolled” out, to say the least.
In this article, we explore the disconnect through the lens of Rogers' Five Factors of Innovation Adoption. In short, the problem isn’t that AI lacks potential; it’s that adoption is held back by mistrust, technical hurdles, and workflow-related barriers. For AI developers, the opportunity is significant if these industry challenges can be overcome, as shown by lessons from technology companies that have successfully closed the gap.
Relative Advantage: AI Doesn’t Clearly Map to the Bottom Line
In iron and steel, the economic drivers underpinning investment are well understood: production output, metal prices and input costs. The challenge for AI is that its capabilities don’t translate directly to these KPIs. Case and point, AISTech had 30 different technical paper topics – from cokemaking to logistics – yet none directly mentioned artificial intelligence. AI is just a means to an end, bundled alongside new instruments, methods and processes.
This puts the onus on AI developers to demonstrate industry-specific ROI over direct investments like equipment and people. As Klaus Stohl (Head of Condition Monitoring and Analysis at Primetals) put it simply: “AI is well-suited for anomaly detection, but explainability to the plant managers is critical for successful implementation.”
One KPI gaining prominence in steelmaking is labor cost, especially as the industry’s aging workforce approaches retirement. Berk Birand (CEO of FeroLabs – AI for industrial process optimization) and Nick Kirkman (Director of Solutions Engineering and Support Ops at Glympse – real-time location intelligence) both highlighted AI’s potential for building and retaining specialized domain knowledge to solve this problem, but only if models are trained on the right data.
Furthermore, large language models (LLMs) still don’t demonstrate a sufficient level of technical performance for users to fully trust them. For instance, Ernie Levinski (Business Development & Decarbonization Leader at Elessent Clean Technologies) occasionally uses Perplexity to research papers and technologies, but as an expert with over 20 years of experience, he still catches it hallucinating technical details.
AI may well add value, but unless that value is expressed in industry-specific metrics and achieves technical performance on par with industry experts, it will continue to struggle to earn investment.
Compatibility: Legacy Workflows Don’t Bend Easily
Modern steelmaking processes are over 100 years old and are rooted in established habits, safety-driven cultures, and passed-down operating procedures. And habits are hard to break.
Pedro Ruiz, CEO of Tamdrea – a pioneer in cloud-driven data and AI for manufacturing and process control – identified a critical challenge: While integrating technology like generative AI into operations offers big advantages, the obstacles to adoption lie in shifting operators away from legacy tools, such as Excel-based data processing, and towards more advanced, cloud-powered AI solutions.
In traditional industries like iron and steel, AI developers must ensure their tools are compatible with the workflows of their end-users, not the buying team (like executives or IT departments).
Complexity: AI Misunderstandings Create Friction
AI is often misunderstood by industry in ways that surprise its developers. "Some people think AI is just ChatGPT," said Ivan Lisboa of Ripik AI (vision AI for anomaly detection). "A major hurdle is that the different types of AI – machine learning for process control, computer vision, LLMs – are often confused. If a client has a bad experience with one, they may lose trust in all the others, even for unrelated use cases."
Even after adoption, understanding how to use AI can be difficult. Tuomas Vuorenmaa (Manager, ICT Engineering at Pesmel – industrial automation systems) observed: “With large language models, people often don’t know how to use them unless someone shows them firsthand.”
For many employees in the iron and steel industry, AI is still a black box. To overcome its complexity, remember that implementation support and education are just as important as technical performance.
Trialability: Consumer AI Use Doesn’t Help Business Adoption
Unlike consumer technology (for instance, ChatGPT), you can’t just swipe a credit card and “try” industrial AI. Deployments often require months of data collection, cleaning, and compliance checks.
Moreover, the idea that consumer adoption of AI will naturally translate into business adoption doesn’t hold up in technical industries like steelmaking. At AISTech, few engineers reported using generative AI for technical tasks, citing a belief that only humans could do their work. One junior engineer said they avoided it entirely, preferring to “use my own brain.” Others limited their use to low-stakes tasks like editing emails or conducting literature reviews.
Worse, poor experiences with consumer chatbots destroy trust for applications in business. As the President of an engineering company stated, “I still make my own presentations because AI-generated content isn’t technical enough and feels like low-quality Google search results.”
IP and privacy concerns further hinder trialability in industry. As a result, many large companies expressed that they were developing internal chatbots, but usage remained low due to limited functionality. This reflects a broad trend of increased disillusionment with generative-AI pilot projects (The Economist).
Trialability for industrial AI remains critical given its novelty. One idea for AI developers is to offer a free version of their applications with constrained functionality, relying on pushed information (e.g., using public information) rather than pulled queries to avoid IP and privacy concerns.
Observability: Demonstrations Over Dashboards
When it comes to building trust, visual proof is paramount. Chris Krechting, Sr. VP of Sales at Everguard AI (computer vision for industrial safety), underscored this: "The only way to get over skepticism is to show it working. Not a dashboard, not a PDF, but something that flags a real issue, with data the operator recognizes."
Silvio dos Santos, Director of Operations at SEMEQ, emphasized that trust from technical teams is earned through clear, evidence-based observation: “When our AI detects an anomaly, our analyst walks the customer through what triggered it. For example, when a foreman improperly lubricated a bearing, our system immediately flagged the issue, and we were able to show the client exactly why the anomaly occurred in the first place.”
Where Progress Is Happening
Despite the frictions, several companies – including those highlighted in this article – are making progress. Successful industrial AI products tend to share four traits:
Human-in-the-loop. As Klaus Stohl noted, “While hallucinations are a concern, generative AI can serve as a great starting point of options for operators to choose from, especially when supported by solid references.”
Domain-specific use cases. Rather than building generic tools, focus on solving tangible challenges that engineers and operators face on the factory floor and can translate to production KPIs. Teams like Fero Labs are taking this approach, tackling well-defined problems such as process optimization with industry-specific data.
Explainability and interpretability. Leverage AI’s ability to synthesize information. As an R&D engineer at an industrial instrumentation company observed, “AI helps clients understand other complex technologies, like radar sensing, faster which promotes adoption.”
Automating text-heavy tasks. As Chidi Aku (Marketing Manager at PSI Software SE, which has developed a conversational operator assistant) stated, “there is a big opportunity to automate low-risk, time-consuming tasks in steelmaking like analyzing operating procedures and manuals”.
Final Thought
The iron and steel sector isn’t resisting AI because it lacks potential. It’s resisting because of legacy workflows, valid skepticism, and poor communication of the domain-specific benefits.
For developers in this space, success will come from doing the slow work of building AI systems that are useable by industry, integrated into existing workflows and more easily adopted. As Noah Smith wrote:
“The future of AI is just going to be like every other technology. There’ll be a giant expensive build-out of infrastructure, followed by a huge bust when people realize they don’t really know how to use AI productively, followed by a slow revival as they figure it out.” (The Economist)
If that slow revival is going to take root in traditional industries like iron and steel, it starts with listening, demonstrating, and building trust – one plant at a time.
About Roundtable
At Roundtable, our mission is to accelerate technology development by equipping R&D teams with purpose-built AI agents that manage projects from idea to experiment. The Roundtable Innovation Agent combines domain-specific AI, results grounded on your company and scientific literature, and automated tools to streamline your work. Serving as a member of your team, our Innovation Agent helps you prioritize, plan, and execute better projects in days, not months, while saving time by handling routine administrative tasks.
Start by booking a demo to explore how our Innovation Agent can support your team today!