The order changes by email and the plan rebuilds itself

The problem

What breaks the plan never arrives structured.

The information that breaks a converting plant's plan almost never arrives structured. It arrives as an email from the bottler pulling four hundred pallets of a water format forward and pushing a soft drink format back. As a notice that the resin tanker is delayed. As a raw material price update. Each of those messages lands in the inbox of someone who cannot watch it in real time. Typical human latency is two to four hours. In that window decisions get made — mounting a mold, starting an order — that then have to be undone. The cost takes two very concrete forms: a mold change that was not needed, with its hours of availability and its kilos to regrind, or a line standing idle waiting for resin that has not arrived. It is exactly the pain any plant describes when it coordinates several lines at once with a high reference mix.

• It arrives as an email pulling in four hundred pallets of a water format and pushing out a soft drink one. As a notice that the resin tanker is delayed. As a raw material price update.
• Each of them lands in the inbox of a person who cannot watch it in real time.
• Typical human latency is two to four hours, and in that window a mold gets mounted and an order started that later has to be undone.
• A converting plant does not sell to consumers: it sells to bottlers and fillers who change the schedule at short notice. That is the norm, not the exception.

How it fits the IRIS system

Connect in chaotic-sources mode — on corporate order channels.

Connect, "chaotic sources" mode. The corporate order inbox and the WhatsApp Business number are silently copied in around the clock. These are corporate ordering channels, not anybody's private conversations. Step by step:

The corporate order mailbox and the company messaging number are order channels, not private conversations. That distinction is what makes the case acceptable to the team and sustainable over time.

• Listening. The message is processed at second zero, without depending on somebody opening the inbox.
• Intent extraction. The model identifies what it is: a pull-forward, a delay, a format change, a non-conformity or a price change.
• Entity extraction. Customer, reference, quantity, committed date, affected batch.
• Cross-reference with live data. Matched against the ERP production schedule and resin coverage in the silos.
• Concrete proposal. iLEAN returns to the planner which line, which mold, in what order and with what coverage, grouping changes by mold family so format changeovers do not multiply.
• Humans in command. The system proposes; the decision and the sign-off stay with the planner.

See the full IRIS architecture →

Before and after

Today's order change versus the cross-checked change

Aspect	Today	With iLEAN Connect
Latency of a critical alert	2-4 hours	Seconds
Basis for the replan	Yesterday's plan	The live data, with silo coverage
Production sequence	Reactive	Grouped by mold and format
Mold changes	Made and unmade	The unnecessary ones avoided
What you receive	One more email	The concrete resequencing of lines and molds
Who decides	The planner	The planner — iLEAN proposes

Impact estimate

Estimated impact — to validate with your own 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 the schedule actually changes.
• The main saving is not the person's time: it is the mold changes avoided.
• Line hours not lost waiting for raw material that was already known to be late.
• From replanning against yesterday's plan to replanning against live data, silo coverage included.

Estimated payback 3-7 months, depending on the real frequency of schedule changes. The main saving is not the person's time but the mold changes avoided and the line hours not lost waiting for raw material. *Estimate to be validated* against the last twelve months of schedule changes.

And the fair question from the production manager

«Are you going to read my sales team's email?» — no. iLEAN sits in copy on the corporate order mailbox and the company messaging number: order channels, not private conversations. On extraction, classifying the intent of an order message is an anchored task and the best models drop below 1.5% error [1] — and the resequencing is executed by a person.

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