The supplier alert that no longer wastes time

The problem

The ingot delay arrives in an inbox nobody reads in real time.

Critical raw-material supplier alerts arrive by email to specific people who can't watch their inbox in real time; by the time they're read, scheduling decisions have already been made and broken, and casting ends up idle waiting for a call.

• Mid-morning, the recycled lead supplier emails a 48-hour delay on the ingot scheduled for that afternoon's casting run. Nothing unusual for a secondary lead smelter, which juggles furnace campaigns and transport of its own.
• It lands in the mailbox of one buyer who is in a meeting, competing with a hundred other messages, written in free text with no fixed format. If that buyer is on leave or on the road, the message may wait even longer.
• By the time it is read, two or three hours later, the casting schedule and the pasting sequence behind it have already been committed. Pasting and curing have their own sequence, so a casting gap ripples down the line for days.
• The grid casting line ends up idle, waiting for someone to call the supplier and decide which alloy or which battery model to bring forward. Meanwhile the alloy that could have been run instead was already scheduled for another day.

How it fits the IRIS system

Connect in chaotic-sources mode — a mailbox in silent copy of every lead supplier, 24/7.

Connect "chaotic sources" mode — a dedicated mailbox (or the supplier's WhatsApp Business line) in silent copy, 24/7. An LLM extracts the intent (supply delay), the material and date, instantly cross-references it with the ERP's active schedule, and returns a concrete action to the right role.

The supplier keeps writing the way it always has. iLEAN reads the intent, the material and the date, checks them against the live casting schedule and tells the casting lead what to move, while there is still time to move it. The difference between a resequence and an idle shift is usually measured in those first two hours.

See the full IRIS architecture →

Before and after

A recycled lead delay today versus read at zero latency

Aspect	Today	With iLEAN Connect
Time until the ingot delay is acted on	2-3 hours of human latency	Seconds
Who learns about it	The buyer, when he opens his inbox	The casting lead, with a concrete action
Check against the casting schedule	Manual, after the fact	Automatic against the ERP's active schedule
Grid casting line during the delay	Idle, waiting for a call	Resequenced to another alloy or model
Supplier channel	Email or WhatsApp to one person	Same channel, read in silent copy
Rescheduling based on	What was read hours earlier	Live data

2-3 hours of human latency → seconds. Rescheduling based on real-time data, not on what was read hours earlier.

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 critical supplier alerts occur.
• From 2-3 hours of human latency to seconds on alerts about recycled lead, oxide, acid or separators. The same mailbox covers every critical supplier, so the value grows with each one added.
• Fewer idle hours on the grid casting line waiting for someone to decide. Pasting and curing keep their sequence instead of absorbing a casting gap.
• And the reschedule is decided on live data, not on a message read too late. We size it with you from the delays logged over the last two quarters.

Estimated payback 3-7 months depending on how often critical supplier alerts occur. *Figure to validate*.

And the fair question from the production manager

“What if it misreads the date or confuses the material?” — extracting intent, material and date from a supplier message is an anchored task, where the best models drop below 1.5% error [1]. And it never reschedules on its own: it proposes the action to the casting lead, who confirms or adjusts it before anything changes in the plan. Every proposal it makes is logged next to the original message.

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