How to Evaluate a Metals AI Vendor: A Buyer's Guide

Most AI vendors selling into manufacturing built their product for a generic factory, then added a metals case study to the website. You can tell the difference in the first ten minutes of a demo, if you know what to ask.

This guide sets out the criteria that separate a metals-specific AI platform from a generic manufacturing tool wearing a metals badge, the red flags to watch for in a vendor demo, and the questions to put to any vendor (GoSmarter included) before you sign.

What to Look For in Metals AI Software

Domain vocabulary, not just industry keywords

A vendor that understands metals uses the terms correctly in conversation, not just on the website. Ask what a heat number is, how it differs from a batch number, and what “3.1 versus 3.2” means under EN 10204. If the answer is vague, the product was built for a different industry.

1D versus 2D cutting scope

Long products (rebar, sections, beams, tube, bar) are a one-dimensional (1D) cutting-stock problem: cutting fixed lengths from stock bars to fulfil orders with minimum scrap. Flat sheet and plate are a two-dimensional (2D) nesting problem: fitting shapes onto a sheet. These are different mathematical problems solved by different tools. A vendor who claims to do both without distinguishing them is oversimplifying, or has not built either properly.

Mill certificate format coverage

Ask how many mill formats the vendor’s tool has actually processed, not how many it claims to support “in theory.” Every steel mill uses its own certificate template, and formats change. A vendor relying on generic optical character recognition (OCR) or a small set of trained templates will need custom work for every new supplier format you throw at it.

Heat and batch traceability, not just stock counts

Generic inventory software counts units in a location. Metals inventory software tracks material by grade, size, condition, heat number, and certificate reference, because two items of the same nominal size can be different materials. Ask to see how the tool handles a bundle of long product that gets split across three separate orders. Does the certificate data follow each piece, or does it stay attached to the original delivery only?

ERP integration approach

Does the tool require you to replace your existing enterprise resource planning (ERP) system, or does it sit alongside it? Ask specifically what integration options exist: CSV export/import, a REST API, or a dedicated connector. “We integrate with everything” is not an answer. “Here is exactly how data flows between our system and yours” is.

Data residency and hosting

If data sovereignty matters to your business or your customers’ contracts, ask where the vendor hosts data and processes it, not just where the company is headquartered. This is often a binary requirement: if the vendor cannot meet it, nothing else about the evaluation matters.

Pricing model: per seat or per site

Per-seat pricing punishes you for growing your team and rewards vendors for keeping usage low. Per-site pricing means the whole team can use the tool without a spreadsheet tracking who has a licence this month. Ask what happens to the price when you add ten more users, not just what the list price is today.

Implementation timeline, honestly stated

Ask for a realistic project plan, not a slide with “weeks” on it. How many of your people are needed, for how many days, and what do they need to prepare? A vendor who cannot answer this in specific terms is guessing, and you will be the one who pays for the guess being wrong.

Human override, not black-box automation

AI that extracts certificate data or generates a cutting plan should show its working and let a person review, adjust, or reject the output before it is acted on. Ask what happens when the AI gets something wrong: does a human see a confidence flag and a clear path to correct it, or is the output trusted blindly?

Audit trail immutability

For compliance-driven industries, an editable log is not an audit trail. Ask whether changes to certificate data, stock records, or cut plans are permanently recorded (who, what, when) and whether that record can be altered after the fact by anyone, including the vendor’s own staff.

Minimum team size and IT dependency

Ask whether the tool needs a project manager, a systems integrator, or your IT department to configure it, or whether a production manager or quality lead can set it up themselves. This tells you whether the vendor is built for operations your size.

Data portability on exit

Ask what happens to your data if you leave. Can you export your stock records, certificate data, and order history in a usable format, or is switching cost part of the vendor’s retention strategy?

GoSmarter vs Generic ERP vs Generic AI Tool

Red Flags in a Metals AI Vendor Demo

“AI-powered” with no specifics. Ask what the AI actually does, what it was trained on, and what happens when it is wrong. A vague answer means the “AI” is a badge on a feature that already existed.

No answer on 1D versus 2D. If a vendor claims to optimise both long-product cutting and sheet nesting without distinguishing the two problems, either the product is thin on both, or the salesperson does not understand what they are selling.

“We work with any mill format” with no evidence. Ask to see a real extraction from an unusual or non-English certificate. If they cannot show you one, the claim is unproven.

Reference customers outside metals. “We have hundreds of manufacturing customers” is not the same as “we have customers who cut rebar or track heat numbers.” Ask for reference customers in your specific niche.

No clear answer on data residency or export. Vendors confident in their product are specific about where your data lives and how you get it back. Vagueness here is usually a sign the answer is inconvenient.

Questions to Ask Before You Sign

1. Can you show me a live extraction from a mill certificate format we actually receive, not a demo template?
2. What does our team need to do differently on day one, and how long before we see a result?
3. What happens to the price when our team grows from five users to fifteen?
4. Where is our data hosted, and what do we get back if we leave?
5. Can you name a customer in metals (not general manufacturing) I can speak to directly?
6. What does the AI do when it is not confident in an extracted value, and who sees that flag?
7. What is genuinely out of scope for your product today, not on the roadmap?

A vendor that answers question seven honestly (GoSmarter’s answer: full manufacturing execution system-style machine scheduling and two-dimensional sheet nesting are both out of scope today) is more trustworthy than one who claims to do everything.

Frequently Asked Questions

Related Resources

• Cloud MES Comparison — how to decide between a full manufacturing execution system and a specialist tool
• GoSmarter vs Generic ERP/MES Planning Modules — a detailed, named comparison against SAP, Dynamics 365, Epicor, and Sage
• Why Some Metals Manufacturers Don’t Choose GoSmarter — an honest look at where GoSmarter is not the right fit
• Mill Certificate Automation: The Complete Guide — how GoSmarter’s MillCert Reader compares to generic OCR and enterprise IDP tools
• Midland Steel Manufacturing Case Study — a real customer’s evaluation and rollout, with results
• GoSmarter Pricing — current tier pricing, per-site model, no per-seat fees
• GoSmarter for Metals Operations — the full platform overview

GoSmarter is made by Nightingale HQ, a UK-based AI company building practical tools for metals manufacturers since 2018.