A supplier alert that doesn't cost 90 minutes of stopped line

The problem

The coil delay arrives in time. Reading it does not.

Critical alerts from steel suppliers or packer clients arrive by email to specific people who can't watch the inbox in real time. By the time they're read, decisions already made get broken — and the lithography line stops for 90 minutes waiting for someone to decide.

• A steel supplier emails a 48-hour delay on a tinplate coil. The message lands in the inbox of one person who cannot watch it in real time.
• By the time it is read, the work order has already been planned around that coil and the litho press is being set up for it.
• The line then stops for 90 minutes waiting for somebody to decide which run to bring forward and whether the right gauge and varnish are in stock.
• Packer clients do the same: a format change or an urgent pull-in written in an email that sits among a hundred other messages.

How it fits the IRIS system

Connect in chaotic-sources mode — a dedicated mailbox in silent copy, 24/7 in metal packaging.

Connect chaotic-sources mode. A dedicated mailbox in silent copy 24/7. An LLM extracts the intent (supply delay, spec change), the entities (supplier, coil, client, date) and cross-checks the ERP, returning the concrete action to the right role.

Finding out sooner is only half of it. The value is that the alert arrives already crossed with the work orders it breaks, addressed to the role that has to decide, with the concrete action on the table.

See the full IRIS architecture →

Before and after

The coil alert in an inbox versus the coil alert crossed with the ERP

Aspect	Today, in the inbox	With iLEAN Connect
Human latency on a critical email	2-3 hours	Seconds
Work orders affected by a late coil	Worked out by hand	Listed with their run and client
Litho line while somebody decides	Stopped up to 90 minutes	Resequenced before it stops
Basis for the new sequence	A message read three hours late	Active work orders, in real time
Who gets it	Whoever the supplier wrote to	Planning, purchasing or quality, by intent
Final decision	Taken late	Taken by a person, on live data

2-3 h of human latency → seconds. Replanning decisions based on real-time data, not on something read three 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 alerts arrive.
• Unplanned stoppages of the litho line caused by missing information, avoided before they start.
• Replanning decided on real-time data instead of on a message read 2-3 hours after it arrived.
• And packer client changes reach planning at the moment they are sent, not when somebody gets back from the floor.

Estimated payback of 3 to 7 months depending on critical alert frequency, avoiding unplanned stoppages from missing information. Estimate to validate.

And the fair question from the production manager

“Will it read everyone's email?” — no. iLEAN sits in copy on a dedicated operational mailbox: it only sees what suppliers and clients send to that channel. Classifying an alert's intent and extracting supplier, coil, client and date is an anchored task where the best models drop below 1.5% error [1], and the replanning is approved by a person before anything moves on the line. If the model is unsure of the intent, the email is escalated as is to a person instead of being classified by guesswork.

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