I spent the better part of thirty years at a steel processing service centre, much of it leading continuous improvement projects with a heavy focus on the shop floor. A good part of that job was investigating promising technology. However, all too often the promising technology was a pre-made solution that turned out to only partially meet our needs, or one that asked us to bend our work processes around it in ways we didn’t want.

So let me begin where I think any honest conversation about bringing new technology into a business, AI included, should start. It is very unlikely that a vendor can walk in cold and tell you what fits your operation. They need to spend time with you first, understanding how you work and your unique challenges. In the absence of me having actually met with you (typically via a Discovery Day engagement), you should treat the posts in this series as descriptive rather than prescriptive. I’m pointing at where AI tends to fit, not telling you what fits your specific situation.

This is the first in a series that works through potential AI use cases one industry at a time. I’m starting with the one I know best (steel service centres), because my experience says there are specific places in a service centre where the work is repetitive and structured enough that AI can carry part of the load. I’m going to highlight five possible solutions; two are things you would buy or configure, and three are things you would build.

The Work That Eats the Week

Anyone who has worked at a service centre knows the daily rhythm of the job. For sales, that is answering the same questions about stock, grades, and order status, over and over. It’s digging a mill certificate or a material specification out of a binder or a shared drive while a customer waits on the line, or it’s turning a request for quote into a real quote against an inventory that has changed multiple times since this morning. For other departments, it is scheduling material through the processing lines, keeping master and pup coil inventories accurate, reviewing secondary offers from the mills, handling claims, or producing the documentation that quality and safety require.

None of that is the craft of the business, it’s just overhead. And overhead that repeats in a predictable shape is exactly where AI can be deployed successfully.

Where I Would Point It

Here are the five places I would consider deploying an AI-integrated solution.

1. Putting your own documents within reach

A service centre sits on a deep pile of its own paper…mill certs, specifications, standard operating procedures, safety and quality records, and decades of history. The answer to most questions is in there somewhere, but it is slow to find. A constrained retrieval tool, which is the category our own Docora product sits in, answers only from the documents you approve and tells you when it doesn’t know rather than inventing something. I wrote about the discipline behind this a few weeks ago, so I’ll keep it brief here. The point worth making here is the deployment speed, since standing up this kind of solution against your own document set can be done very quickly.

A few concrete places this could go to work:

  • Policy documents. A knowledge base of policies and previously published clarifications, made available to shop-floor supervisors so they can confidently answer questions from their reports, especially on off-shifts when nobody else is around.
  • Maintenance history. A knowledge base of reported maintenance issues and the completed work orders, so employees can quickly see when similar issues came up before and what was done to correct them.
  • Mill specs. A knowledge base of every supplying mill’s specifications, so sales reps can quickly query which mill or mills produce steel meeting a particular set of characteristics.

2. Turning requests for quote into quotes

In most service centres, quoting can be a very time-consuming art. The incoming requests arrive in all sorts of formats, from a detailed spreadsheet to a one-line email. They often need to be interpreted or have the customer’s spec matched to a mill’s specific capabilities. This is where an AI-integrated solution can serve as the front end, reading the incoming request in whatever format it arrives in, cross-referencing it against OEM specs, mill capabilities, inventory, and previous orders, and preparing a draft response in a standard format using your own pricing logic. The system takes care of the tedious prep work, and the salesperson enters the process with a draft in hand that they can review and adjust using their own experience and judgment.

This is where a custom build is required to reach the level of integration and customization that makes the process fit your operation. But with a process built around your logic and connected to your organization’s knowledge, the salesperson gets to spend their time on high-value work.

3. A CRM system built around how you actually sell

Most off-the-shelf customer relationship management (CRM) systems assume a sales flow a service centre does not mirror. Your book of business is relationships, repeat tonnage, volume discounts (on both the buy and sell side), and quote history. The honest version is a system built around the way you really sell, with AI inside it to prospect for customers, rank and enrich contacts, and to run analysis across your order book. AI is especially effective at surfacing patterns in your existing book of business that go well beyond simple buying-pattern analysis. This is custom development, and it’s one of the things we build, because off-the-shelf tools are just not specific enough.

4. Working with raw and finished goods

This is the big one for metal processing, and the area where the real dollars live. Coil and inventory movement, cut and nest sequencing, getting honest use out of pup coils and offcuts, and scheduling material through the processing lines are all problems with significant variation, but also with enough structure to model and optimize with AI.

A concrete place where AI could be used is an AI agent that actively monitors your inventory, looking for opportunities to offer off-cuts or pup coils to suitable customers or prospects, or even recommending that some of them be scrapped.

I’ll be straight with you, engagements in this area are heavier than the others. These are custom builds, scoped to the way your operation runs and integrated with your ERP, and not something you just install and switch on. However, it’s also the area where material yield and line utilization can more than justify the effort.

5. Answering the phone

Although a lot of customer interaction now comes through email and text, a service centre still fields plenty of phone calls every day. Prospective customers call with inquiries, and existing customers call about order status, stock and availability, and shipment updates. These are exactly the kind of calls an AI voice system handles well when the integration underneath is done properly. I covered this one in detail last week, so I will simply point you back to that piece.

Where AI Is the Wrong Tool on This Floor

AI is not always the answer, and there are places in a steel service centre where it does not belong. Anything safety-critical or metallurgical that needs a qualified human sign-off stays with the qualified human — although the AI can often do some of the prep work. The judgment calls a good salesperson or operations lead makes by reading a customer or a job stay with that person. And of course, any problem that is really a process or a data problem in disguise will not be solved by putting AI on top of it. If your inventory records are wrong today, AI will simply hand you wrong answers faster.

There is a sensible order to all this as well. The document retrieval and the phone handling are the low-risk places to begin. Quoting sits in the middle, a real build but a contained one, and often the most keenly felt, because everyone involved knows how much work quoting takes. The customer system and the material management workflows are larger commitments, and they should follow a clear-eyed look at whether your data and your processes are ready to be built on.

The First Step

I’ll end where I started: I can’t tell you from the outside which of these fits your operation, or in what order they should come. That’s what a Discovery Day engagement is for, where I spend a day with your team to surface the first places where AI could start taking on real work, and, just as importantly, to tell you honestly where it would not.

Next time I’ll look at some AI use cases in a different industry, again asking where AI fits and where it doesn’t.