At the last MSCI annual meeting, the word "AI" appeared in 23 different vendor booth displays. By our count, exactly four of those vendors could demonstrate a working AI feature that did something useful for a steel service center. The rest had slide decks.
This is not a criticism of ambition. AI will transform steel distribution. But the gap between what vendors promise and what they deliver is wide enough to drive a flatbed truck through. Service center operators deserve an honest accounting of where AI stands today, what it can realistically do, and where the hype outpaces the technology.
What AI Can Do Today
Automated Quote Generation
This is the most immediately valuable AI application in steel distribution. A sales rep receives an RFQ by email. The AI reads the request, identifies the material specs (grade, gauge, width, length, quantity), checks real-time inventory across all warehouse locations, applies the correct pricing (base metal plus extras, processing charges, freight estimates), factors in the customer's pricing history and margin targets, and generates a formatted quote document.
What used to take 30 to 45 minutes takes under 5. The AI does not replace the sales rep's judgment. The rep reviews the quote, adjusts if needed, and sends it. But the grunt work of pulling data from four different screens, running calculations, and formatting the document is gone.
Intelligent Search and Data Retrieval
Steel service centers generate enormous amounts of data: orders, quotes, MTRs, inspection records, invoices, shipping documents, customer communications. Finding a specific piece of information traditionally means knowing which system to look in and how to query it.
AI changes this. A warehouse manager can ask, in plain English, "Show me all open orders for Johnson Controls that include 14-gauge galvanized" and get an answer in seconds. A credit manager can ask "Which customers have gone over 60 days in the last quarter?" without building a report. The AI understands context, handles fuzzy matching, and searches across the entire system.
Pricing Recommendations
Steel pricing is part science, part art. The science is market data: HRC indices, scrap prices, import levels, demand indicators. The art is knowing the customer, reading the competitive situation, and deciding when to hold margin versus when to sharpen the pencil.
AI handles the science exceptionally well. It can process market data feeds, analyze the customer's historical purchasing patterns, compare your pricing to recent wins and losses, and recommend a price point that balances margin protection with win probability. The art still belongs to the sales rep. But making pricing decisions without the science is like navigating without a map.
What AI Cannot Do Yet
Fully Autonomous Purchasing Decisions
Some vendors suggest that AI can manage your purchasing: automatically placing orders with mills when inventory drops below threshold, optimizing buy quantities based on demand forecasts, timing purchases to market cycles. The concept is sound. The execution is not ready.
Steel purchasing involves relationships, lead times that vary by mill and product, quality considerations that change by application, and market timing decisions that depend on information AI cannot access (conversations with mill reps, rumors about production curtailments, customer pipeline that has not yet turned into orders). Fully autonomous purchasing is a five-year-plus proposition. Today, AI should inform purchasing decisions, not make them.
Real-Time Yield Optimization
The idea of AI monitoring a slitting line or shear and making real-time adjustments to optimize yield is technically fascinating and practically premature. It requires sensor integration, machine control interfaces, and a feedback loop that operates in milliseconds. Some large mills are experimenting with this. Service centers with processing equipment from the 1990s and 2000s are not there yet.
Where AI does help with yield today: analyzing historical production data to identify patterns. Which jobs consistently produce excess scrap? Which operators achieve better yield on which machines? Which material sources have higher defect rates? This is retrospective analysis, not real-time optimization, but it drives real improvement.
How to Evaluate AI Claims
When a vendor says "AI-powered," ask three questions:
First: Can you show me it working with real steel data? A demo with fake data proves nothing. The AI should handle actual material descriptions, real pricing scenarios, and genuine customer queries. If the demo only works with pre-scripted examples, it is a prototype, not a product.
Second: What happens when the AI is wrong? Every AI system makes mistakes. The question is whether the system is designed to fail safely. Does a wrong price recommendation go straight to the customer, or does a human review it? Can users correct the AI, and does it learn from corrections?
Third: Where is my data going? AI models need data to work. Some vendors send your pricing data, customer information, and order history to third-party AI providers. You should know exactly where your data goes, who can access it, and whether it is used to train models that benefit your competitors.
The Honest Timeline
AI in steel distribution is real, valuable, and early. The companies adopting it now are building a compounding advantage: better data, faster decisions, more capacity freed from administrative work. The companies waiting for AI to be "proven" will find themselves playing catch-up against competitors who have been training their systems on a year's worth of operational data.
Start with the applications that work today. Be skeptical of promises about tomorrow. And remember that the best AI in the world is useless if it is not built on software that understands how a steel service center actually operates.