The supplier's alert, in seconds, not hours

The problem

The delay notice is read when the line is already waiting for the material.

Critical supplier alerts arrive by email to people who can't watch the inbox in real time. By the time they're read, planning decisions are already broken — and on a multi-line continuous process, that break spreads.

• Clay, feldspar, sand and glaze frits arrive by bulk truck or big bag, and the suppliers warn of delays the way they always have: by email, to the buyer's personal inbox.
• The buyer is not watching the inbox in real time. The notice is read hours later, often when a line is already scheduled to start with that material and the mill or the glaze preparation is already committed. Nobody is at fault; the channel was simply never designed for urgency.
• Mill feed recipes and glaze preparation are planned days ahead. A frit delay discovered on the day of the startup leaves a glazing line without its glaze and a press without its next format.
• And in a continuous multi-line process the break does not stay local: the atomizer, the press, the glazing line and the kiln all lose their sequence at once, and the kiln cannot simply be paused. What was a supply problem becomes a production problem on six lines.

How it fits the IRIS system

Connect in chaotic-sources mode — the supplier inbox, read 24/7 in silent copy.

Connect chaotic-sources mode — a mailbox in silent 24/7 copy. A language model extracts intent (supply delay) and entities (supplier, material, dates), cross-references the ERP, and returns the concrete action to the right role.

The supplier keeps writing to the same address and the buyer keeps reading the same inbox. What changes is that a silent copy is read the moment it lands, and the planner gets the affected line and the concrete action, not just a forwarded email. The decision is still theirs; it is simply made hours earlier. In a process that cannot stop cheaply, those hours are the difference between a sequence change and a stopped line.

See the full IRIS architecture →

Before and after

The supplier email today versus the email read on arrival

Aspect	Today	With iLEAN Connect
Time until the delay is known	Hours, when someone opens the inbox	Seconds
Material and quantity affected	Read and interpreted by hand	Extracted from the email text
Which line and startup it hits	Worked out in a meeting	Cross-referenced with the ERP
Glaze preparation for the next color	Discovered short of frit	Replanned in advance
Who receives the action	Whoever was copied	The role that has to decide
Supplier's way of writing	—	Unchanged, any format

Hours of human latency → seconds. Replanning decisions based on the lines' real-time state, not on an email read hours later.

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 3-7 months, depending on how often supplier alerts arrive.
• Unplanned stops avoided when the cause is a notice that was read too late, not a material that truly could not be replaced.
• From hours of human latency to seconds between the supplier's email and the planner's decision. The planner on duty sees the alert with the affected line already named.
• Replanning based on the real state of the lines, not on an email read after the sequence has already broken.

Avoided unplanned stops from miscommunication, estimated payback 3-7 months depending on alert frequency. *Estimate to validate*.

And the fair question from the production manager

“Can it tell a delay notice from an order confirmation?” — classifying the intent of a supplier email and pulling out material, quantity and dates is an anchored task, where the best models drop below 1.5% error [1]. And nothing is replanned automatically: the suggested action goes to a planner, who confirms or discards it, and that decision is recorded. Over time, that record shows which suppliers and materials cause most replanning.

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