The pallet is labeled from the order, with no keying in between

The problem

The failure point with the highest consequence per unit of effort.

It is the classic "almost integrated" system. The pallet print-and-apply unit is five or ten years old, has a command interface or an API, and remains disconnected because integrating it properly never becomes the priority over what is urgent. The result is manual keying into the machine panel at every batch change. It is a ten second gesture, and it is the failure point with the greatest consequence per unit of effort in the entire plant. One wrong digit and a whole pallet reaches the customer under the wrong batch. Best case, a return with its transport and its difficult conversation. Worst case, broken traceability: when a quality problem appears you cannot bound which production is affected, and you end up blocking more days than necessary out of caution.

• The pallet label is the last information leaving the plant and the first the bottler reads: part number, lot, mold, count, destination customer and declared recycled content.
• The print-and-apply unit is five or ten years old, has an API, and is still disconnected because integrating it properly never rises above the urgent.
• The result is manual keying on the machine panel at every lot change. It is a ten-second gesture.
• One wrong digit and an entire pallet reaches the customer under the wrong lot — and with it, traceability broken at exactly the point the customer looks at first.

How it fits the IRIS system

Connect as stitching between two modern systems, through APIs.

Connect as a stitch between two modern systems, over the API. Neither one is replaced. Step by step:

The loop closes with vision: the in-line camera reads the printed label and checks it against the live order. If they disagree, JIDOKA AI stops. Publishing the right data and verifying it was printed are two different things, and here both are covered.

• Trigger from the ERP. When the ERP marks an order as started, it publishes the complete data set.
• Publication to the labeller. Reference, batch, mold and cavities, count per pallet, destination customer and declared recycled content travel to the equipment over the API.
• Printing. The GS1-128 label prints from source data, with no manual intervention. Changing batch becomes a tap in the ERP.
• Closing the loop with Edge. The inline vision camera reads the printed label and checks it against the live order; if they do not match, JIDOKA AI stops the line before the pallet leaves the plant.
• Complete traceability. The pallet is linked to the mold, the resin batch and the lab tests for the period.

See the full IRIS architecture →

Before and after

The isolated labeler versus the stitched labeler

Aspect	Today	With iLEAN Connect
Keystrokes per lot change	One on the machine panel	Zero
Typing errors	Several a month	Zero of human origin
Returns from mislabeled pallets	Recurring	Sharply reduced
Bounding a quality problem	Block as a precaution	Bound by real traceability
Declared recycled content	Typed	Travels from the order
Verification of what was printed	None	The camera checks it against the order

Impact estimate

Estimated impact — to validate with your own 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 — among the fastest in the pool, because the machine already ships an API and the integration effort is low.
• Returns avoided from mislabeled pallets.
• Production days that stop being blocked as a precaution when a quality problem appears.
• From several typing errors a month to zero of human origin.

Estimated payback 3-6 months — among the fastest in the pool, because the integration effort is low (the machine already has an API) and the risk it removes is high. The return concentrates in avoided returns and in production days that stop being blocked out of caution. *Estimate to be validated* against the history of returns and labeling incidents. *Technical note*: if the installed equipment does not offer an open API, the case still holds but requires an intermediate adapter, which raises initial CAPEX somewhat.

And the fair question from the production manager

«If the machine already has an API, why don't we integrate it ourselves?» — you can, and that is the right answer the day the project reaches the top of the list. The problem is that it has not for years, because it competes with the urgent. This stitching replaces neither system and asks for no shutdown window, so it does not join that queue.

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