The bundle label is your product. And today it is produced by typing

The problem

The bundle label is your product, and today it is produced by typing.

Somebody weighs the bundle, reads the number off the display, types it into the labeling system together with the order and the heat, and prints. One wrong digit in the weight or the heat turns a correct bundle into unidentifiable material. And a correct bundle with the wrong label is, in practice, rejected material: the customer does not argue about what the steel is, but about what the label says. What is striking is that both machines are relatively modern and already expose an API or a data port. They are isolated not because it is technically impossible, but because integrating them properly takes time and is never the priority of the quarter. The classic "almost integrated, but not".

• Somebody weighs the bundle, reads the number off the display, types it into the labeling system together with the order and the heat, and prints.
• One wrong digit in the weight or the heat turns a correct bundle into unidentifiable material. The customer does not argue about what the steel is, but about what the label says.
• Both machines are relatively modern and already expose an API or a data port. They are isolated not because it is technically impossible, but because integrating them was never the priority of the quarter.
• Meanwhile the real weight of every bundle — the number that would give coil yield automatically — is typed into a label and never reaches anything else.

How it fits the IRIS system

Connect stitches scale and labeler to the ERP through their APIs, replacing neither.

Connect stitches two modern systems through the API, replacing neither. With the cutting run active, the ERP publishes order, heat, grade, coating and dimensions; the scale publishes the real bundle weight; the labeler prints the combination. Changing the run means a change in the ERP, not typing on the labeler panel.

With the cutting run active, the ERP publishes order, heat, grade, coating and dimensions; the scale publishes the real weight; the labeler prints the combination. The label that goes on is the right one by construction, and as a side effect the real weight of every bundle turns coil yield into a number nobody balances by hand.

See the full IRIS architecture →

Before and after

The isolated scale and labeler versus the stitched ones

Aspect	Today	With iLEAN Connect
Weight on the label	Read off the display and typed	Published by the scale
Heat and order on the label	Typed with the weight	Published by the ERP from the active run
Labeling errors	Several a month, some with a claim	Zero human error at source
Changing the run	Typing on the labeler panel	A change in the ERP
Coil yield	A spreadsheet balanced by hand	Calculated from real bundle weights
Either machine	—	Neither is replaced

from several labeling errors a month — some with a claim attached — to zero human error at source in identification. Plus a side effect worth the project on its own: by recording the real weight of every bundle, coil yield is calculated automatically, without the spreadsheet somebody balances by hand today.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 3-6 months — the fastest in this matrix.
• A sharp reduction in claim risk from incorrect identification, which is the claim where the customer is always right.
• A yield-per-coil metric that appears automatically from the real weight of every bundle, without the spreadsheet somebody balances by hand today.
• And the changeover on the labeler disappears: changing the run is a change in the ERP, not typing on a panel.

fast payback, estimated 3-6 months, with a sharp reduction in claim risk from incorrect identification and a yield-per-coil metric appearing that today is calculated by hand. *An estimate to be validated*.

And the fair question from the production manager

“Isn't a classic integration cleaner?” — it is, the day the project reaches the top of the list, and that is the problem: this is the textbook case of two modern machines still isolated because there was never time. Here the identity on the label comes from the ERP through the API, not from a model reading anything, so the question of model error does not even arise at that point. Where a model does step in — matching a customer's free-text labeling requirement to the fields the ERP publishes — it is an anchored task where the best models drop below 1.5% error [1], and the label template is validated once per customer, not per bundle.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.