Zero Line Stoppages From Supply Shortfalls

The problem

The line does not stop when the part is late. It stops when nobody saw it coming.

An unanticipated line stoppage from a supply shortfall is the biggest recurring economic risk for a plant relying on hundreds of local and international just-in-time suppliers. Each incident costs anywhere from tens to hundreds of thousands of dollars in downtime, plus the risk of missing market volume commitments.

• A plant fed by hundreds of local and international JIT suppliers lives with one recurring risk above all others: the unanticipated line stoppage because a part is not at the station.
• Each incident costs from tens to hundreds of thousands of dollars in downtime, plus the risk of missing the volume committed to the market.
• The signals that would have warned in time usually existed: a supplier email, a truck that did not arrive, a container that never reached the line side. They lived in different systems and different heads.
• So the reaction starts when the andon is already pulled, and the only options left are an emergency shipment or a stopped line.

How it fits the IRIS system

Connect, Edge and Agents in cross-checking rings — before the line stops.

Connect (gaps 3 and 7) detects the early supplier signal and confirms real physical arrival; Edge plus legacy integration (gaps 8 and 10) confirm the state of the critical point depending on that part; Agents (gap 11) keeps an auditable record of every decision; SMED AI resequences unaffected units and JIDOKA AI blocks when needed — all before the line has to stop.

No single case prevents the stoppage on its own: the supplier alert, the dock receipt and the line-side check each see one part of the picture. The value is in crossing them, and in letting SMED AI resequence the units that do not need the missing part while JIDOKA AI holds the ones that do. The line keeps moving with the units it can build, and the gap is filled when the part arrives.

See the full IRIS architecture →

Before and after

The rings of the system and what each one confirms

Ring	Which cases it draws on	What it confirms
1 · Early signal	Supplier delay alert (3)	The part will be late, and for which VINs
2 · Physical arrival	JIT receiving at the dock (7)	The part is really in the plant
3 · Critical point	Torque tool and MES (8)	The station can build that VIN correctly
4 · Paint sequence	Color change validation (10)	The body is ready in the right color and order
5 · Evidence	IATF evidence pack (11)	Every decision is recorded and auditable
6 · Response	SMED AI and JIDOKA AI	Unaffected units resequenced, affected ones held

Reactive line stoppage once the part is already missing at the station → proactive resequencing minutes or hours before the line needs it.

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, measured against the first line stoppage avoided, to be validated with materials planning and plant management.
• Each avoided stoppage protects tens to hundreds of thousands of dollars in downtime.
• From a reactive stoppage when the part is already missing at the station to proactive resequencing minutes or hours before the line needs it.
• And protection of the volume committed to the market, which is what plant management really answers for.

Estimated payback 6-12 months against the first stoppage avoided. *To be validated* with materials planning and plant management. (Payback 6-12 months · protects against the cost of downtime)

And the fair question from the production manager

“Do we need all twelve cases running before this works?” — no. The rings are built in layers, and the supplier alert plus the dock receipt already give you the first two. Crossing a delay warning with the live sequence is an anchored task, where the best models drop below 1.5% error [1], and no resequencing happens without a planner confirming it.

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