Kanban across plants and countries — customs never notifies the MRP.

The problem

A kanban with a constant lead time, on a chain that no longer has one.

The planning manager of a group with plants in several countries knows kanban works — it is one of the best methods industry has produced. What no longer works is the assumption it was calculated on: that the lead time between the plant producing the component and the plant assembling it is stable and known.

1. The real transit time is not in the MRP — it is in the freight forwarder's email, in the carrier's messaging app, on the customs portal, in the head of the logistics manager who calls every morning to ask.
2. The safety buffer was set once and never touched — because touching it is frightening. The result: either stock piles up and ties down millions, or material is missing on exactly the day it matters.
3. Plants don't talk to each other — each has its own ERP (sometimes three different ERPs inside the group), its own material codes, its own calendars. Coordination happens through weekly meetings and spreadsheets that age badly.

The paradox of multi-plant multi-country kanban: the cleanest method in Lean runs on the dirtiest data, because the good data sits outside the system.

How it fits the IRIS system

iLEAN doesn't replace the MRP — it seals the cracks between planning, transport and customs.

The IRIS category (Industrial Reality Intelligence Systems) starts from one principle: capture reality first, produce the report afterwards. In cross-plant kanban that reality almost never arrives through a sensor — it arrives in a forwarder's email or a carrier's message. Connect captures that flow and turns it into data. iLEAN acts as filler between each plant's MRP/ERP and the outside world, without asking you to swap one ERP for another.

Connect captures transport notices wherever they land. The agent cross-checks them against each plant's plan and recalculates the kanban. The manager signs off — the system warns before the impact, not after.

The three iLEAN pieces applied to multi-plant multi-country kanban:

• Brain — collects each plant's data (real consumption, defect ratios, calendars) and keeps a unified view of the dynamic kanban. It is not a BI: it is a layer that updates reorder points per corridor as soon as the real lead time changes.
• Connect — reads the forwarder's emails, the carrier's messages, the customs portal, the transcribed call from the broker. And it captures from each plant's ERP/MRP through whichever door each one allows (API, ODBC, export). Bidirectional: it drafts the reply to the carrier on the same channel, and the person signs it off.
• Agent — cross-checks the delay notice with the destination plant's plan. If there is a risk of a stockout, it proposes redirecting from a sister plant, pulling the next kanban forward, or warning the line to reorganize the day. The planning manager approves; the agent executes only where explicit autonomy has been granted.

See the full IRIS architecture →

Before and after

Classic cross-plant kanban vs. dynamic kanban with iLEAN

Aspect	Classic cross-plant kanban	With iLEAN Brain + Connect + Agent
Lead time	Constant, set years ago	Variable, calculated from the corridor's real history
Customs notices	Forwarder email someone re-sends at 10 pm	Captured at second zero, agent recalculates the impact
Multi-ERP	Each plant on its island, coordination by weekly meeting	Connect reads every ERP, agent synchronizes catalogues
Safety stock	Constant buffer nobody dares to touch	Dynamic buffer per corridor, tuned to the real risk
Reaction to an event (strike, hold-up)	Crisis meeting the next day	Redirection proposal from the sister plant before the impact
Visibility for the line	The operator finds out when material is missing	The line sees the adjustment coming hours or days ahead

Impact estimate

Impact estimate for your plant — to be validated with your own numbers.

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

• Industrial group with 3-5 plants in different countries, multi-ERP, and a plant-to-plant lead time crossing at least one customs border.
• Brain + Connect + Agent pilot on one critical corridor (the pair of plants with the highest flow or the most stockouts). First expected value in a few weeks: unified capture of transport notices alone removes the manual forwarding and gives day-by-day visibility.
• Reduction of cross-plant safety stock in the order of 15%-30% in the first year (estimate to be validated). Reduction of downtime caused by missing components, of a similar order.
• Indicative payback between 4 and 9 months. The hard lever is twofold: working capital released + line stoppages avoided at the destination plant.

And the operations director's reasonable doubt

“What if the agent redirects a kanban badly and leaves me without material in another plant?” — hallucination is a problem of free generation, not of anchored tasks. On tasks where the AI simply recontextualizes a piece of data (reading a delay notice and cross-checking it with the plan), the best models brought the error rate below 1.5% ^[1]. And even so, the agent proposes, the planning manager decides. The three safety rings let each corridor be configured with the level of autonomy the group defines, without touching code.

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks. ^[2] Ocean freight variability data: ~1,500 USD per container up to >20,000 at the 2021 peak, a 2024 rebound driven by the Red Sea, then a fall from overcapacity.