From the supplier lot to the exact unit, in minutes

The problem

A defective supplier lot is a question nobody can answer in time.

When a supplier lot — compressor, control board, foam, sheet metal — turns out defective or gets a safety alert, you need to narrow down in minutes, not weeks, which finished-unit serial numbers carry it and which countries they're in. That reconstruction today depends on manually cross-referencing receiving slips, build sheets and final-test records from different sources — risking hundreds or thousands of units already shipped to dozens of countries with no clear traceability.

• When a supplier lot — compressor, control board, foam, sheet metal — turns out defective or gets a safety alert, the question is immediate: which finished coolers carry it, and in which countries are they?
• Today the answer depends on cross-referencing receiving slips, build sheets and final-test records by hand, from different sources. Each source was built by a different department and none of them shares a key with the others.
• That takes weeks, and meanwhile hundreds or thousands of units are already shipped to dozens of countries.
• Without clear traceability, the only safe scope is a wide one: far more units recalled or inspected than actually carry the suspect lot. And the bottler or retail customer, who has the coolers in their stores, wants an answer before the plant can give one.

How it fits the IRIS system

Four coordinated traceability rings — the previous cases, working as one system.

The iLEAN flagship system coordinates 4 rings: the supplier lot captured at receiving, the lot consumed by the assembly production order, the serial number engraved by the serializer, and the final test result with its auditable dossier. If the 4 rings don't match for a unit, JIDOKA AI flags it before it leaves the plant; SMED AI speeds up export configuration changeovers so the coordination doesn't cost productivity.

No single ring answers the question; each already exists in some form and the answer still takes weeks. The value is in linking them: the supplier lot at receiving, the order that consumed it, the serial number engraved and the test that cleared the unit. When the four agree, a unit's history is complete; when they disagree, it is caught before the unit leaves the plant.

See the full IRIS architecture →

Before and after

The four rings, and what each one guarantees

Ring	Which cases it draws on	What it guarantees
1 · Receiving	Incoming slip (7) + supplier alert (3)	The supplier lot enters correctly
2 · Assembly	Line startup (1) + foam panel (2) + headset (4)	Which order consumed which lot
3 · Serialization	Serializer (8) + export voltage (10)	One serial number per cooler, same in plant and ERP
4 · Final test	Test tablet (5) + lab (6) + dossier (11)	The test that cleared the unit
Cross-check in commercial refrigeration	JIDOKA AI	A unit whose rings disagree does not leave
Line time	SMED AI	Export changeovers stay fast

Weeks of manual rebuilding after a supplier alert → a minutes-long lookup by lot or serial number.

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 6-12 months against the first avoided mass-traceability event, to validate with Quality, Supplier Development and General Management.
• Weeks of manual rebuilding after a supplier alert become a minutes-long lookup by lot or by serial number. The answer to a supplier alert stops depending on who is available to search paper that week.
• The scope of any field action narrows from everything built in a period to the exact coolers that carry the suspect lot.
• And the plant can answer a bottler or retail customer with serial numbers and countries, not estimates. Supplier development can also hold the component supplier to account with exact figures.

Estimated payback of 6-12 months against the first avoided mass-traceability event. Estimate to validate with Quality, Supplier Development and General Management.

And the fair question from the production manager

“Do we need all twelve cases before this works?” — no, and framing it that way would mean never starting. The rings are built layer by layer, each case pays on its own, and receiving plus serialization already give the first lookups. Linking records across rings is an anchored task, where the best models drop below 1.5% error [1]; when two rings disagree, JIDOKA AI holds the unit and a person decides.

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