Zero PPM to the Tier-1, with the IATF evidence alive every day

The problem

A rising PPM in the quarterly report is not an ugly number: it is your place on the supplier panel at stake.

In a Tier-2 automotive stamper, a rising PPM in the quarterly report is the biggest reputational and economic risk in the sub-sector. Every container returned by the Tier-1 triggers a full 8D, countermeasures, the process lead's time spent on containment and the risk of a hold on the part number — which freezes billing for that reference until the customer lifts it. But the direct cost is the least of it: behind it comes the damage to the relationship with the buyer — built approval by approval over years — and the risk that a rising PPM closes the door to new parts and pushes you out of the business.

And the nature of the process multiplies the risk: every coil comes from a different heat, die changeovers are frequent and cadence pressure pushes towards starting fast. Between uncoiling and the shipped container there are enough strokes and enough hands for memory not to be an acceptable protection mechanism. Either the four points verify each other on the shop floor, or the customer makes the first finding and it gets reported as PPM.

1. Ring 1 · ERP and coil. The active production order defines the part number, the lot and the mounted coil, and iLEAN cross-references it against that coil's mill certificate as the single source of truth. If the mounted material does not match what the order requires, it fires before the first stroke. Everything that happens afterwards is measured against this definition, not against anyone's memory.
2. Ring 2 · Press curve. The camera on the HMI feeds the system with peak force and strokes per minute in real time; any drift from the active reference pattern raises an alert. Every stroke's force signature stops being a figure only the operator looks at and becomes a witness cross-checked against the order and against what Edge sees.
3. Ring 3 · Edge on the line. The overhead and side cameras detect the defect part by part, and at an order change Edge validates the die changeover against the active order. If something does not add up — a part outside the pattern, a die that is not the one the order requires — JIDOKA AI blocks the line in that instant: the discrepancy is stopped on the shop floor, not in the container.
4. Ring 4 · IATF evidence pack. Every lot ships with an auditable dossier cross-referencing the three previous rings — the order and the coil that defined it, the press curve that produced it and the Edge that verified it — plus first-off, last-off and CMM. Full traceability runs from coil → heat → press → die → part → container → customer. If the auditor or the buyer asks, the answer already exists reconciled; nobody has to rebuild it by hand.

A fifth element is not a ring, it is what keeps the four from penalizing productivity: SMED AI accelerates the die changeover — with the propagation of part number, coil, target curve and labeling included — so that coordinating four verification points does not mean sacrificing OEE. Coordination protects PPM without slowing anything down.

See the full IRIS architecture →

Satellite cases from the same plant, with the technical detail of each ring: the order and the coil cross-referenced against the mill certificate (Ring 1) · the press curve from the panel in real time (Ring 2) · part-by-part defect detection with Edge (Ring 3) · die changeover validation with Edge (Ring 3) · evidence pack for IATF (Ring 4).

Before and after

A plant with no coordinated rings vs. a plant with the 4 iLEAN rings

Aspect	Plant with no coordination	Plant with the 4 iLEAN rings
Production order and mounted coil (part number, lot, mill certificate)	Read on paper or screen and checked by hand; the mill certificate lives in a PDF nobody cross-references stroke by stroke	Ring 1 (ERP): the active order defines part number, lot and coil as the single source, cross-referenced against the mill certificate before the first stroke
Press curve (peak force and strokes/min)	Only the operator looks at the force signature on the HMI; drift is discovered once the part has already left	Ring 2: the camera on the HMI streams peak force and strokes/min in real time; any drift raises an alert cross-checked against the order
Part defect and die changeover	Manual visual sampling; the wrong die or the non-conforming part is discovered at the Tier-1 receiving	Ring 3 (Edge): overhead and side cameras inspect part by part and validate the die changeover; JIDOKA AI blocks the line if anything does not match
Evidence and IATF audit per lot	Reactive IATF audit; evidence rebuilt by hand when the customer complains or the auditor arrives	Ring 4: dossier ready every day, cross-referencing the three previous rings plus first-off, last-off and CMM, signed by the quality technician
Die changeover and OEE	Penalizes OEE; changeover pressure multiplies dirty starts and non-conforming first parts	SMED AI accelerates the change with order, coil, curve and labeling propagation included; fewer dirty starts
PPM to the Tier-1 with no active control	Rising PPM in the quarterly report, some with a risk of a part number hold	Zero — detection on the line (on the shop floor), not at the customer; no container of non-conforming parts leaves for the Tier-1

Impact estimate

Impact estimate for your plant — to be validated 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.

• Tier-2 stamper with several presses and part families, seating and suspension Tier-1 customers with IATF approval, coils from different heats and frequent die changeovers.
• Pilot on the highest-risk press with the 4 rings coordinated — order and coil, press curve, Edge with JIDOKA AI and the IATF evidence pack. First value within a few weeks, starting with the most urgent ring.
• Quality's hard lever is the protection of the IATF approval and of the multi-year relationship with the Tier-1 — an asset whose value far exceeds the one-off cost of any single PPM incident.
• Operations' hard lever is avoiding the 8D, the containment and the risk of a part number hold that a single non-conforming container triggers, plus the OEE recovered by avoiding dirty starts after the die changeover.
• Indicative payback of 6 to 12 months, counted against the first PPM incident avoided. It also enables the move into active-safety critical parts and direct export to OEMs. With a single serious event avoided, the system pays for itself.

And the fair question: "what if the system itself gets it wrong?"

The right question from the quality director. Two answers that hold each other up. The technical one: hallucination is a problem of free generation, not of anchored tasks — cross-referencing a coil's mill certificate against the order's part number, reading peak force and strokes/min on the press curve and comparing them with the active reference pattern, or verifying that the part Edge sees matches what the other three rings say, are anchored tasks of the clearest kind. In tasks of this type, the best models brought the error below 1.5% ^[1]. The architectural one: the system is built on the three IRIS safety rings — Connect carries, Agents decide, the person signs. And because the 4 rings of this case cross-check each other, an isolated failure in one is exposed by the other three before it becomes a reported PPM. It is not that everything goes through a person; it is that the system allows it where it matters, and here it matters.

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