EDI brings the schedule; email brings reality

The problem

EDI covers the schedule; it does not cover the exceptional, which is what costs money.

EDI covers schedules and firm calls. It does not cover the exceptional, and the exceptional is what costs money. An alert read two hours late means material built in the wrong order, a rack that has to be broken down, or a special transport paid for. The cost of this appears in no budget line: it dissolves into urgent freight, expedition overtime and rack rework. That is precisely why it is never tackled — nobody knows what it adds up to in a year. And it is not solved by putting more people on the mailbox: it is solved by removing the latency between the message arriving and somebody with decision authority knowing what to do.

• An alert read two hours late means material built in the wrong order, a rack to break down, or a special transport paid for.
• The cost appears in no budget line: it dissolves into urgent freight, expedition overtime and rack rework. That is why it is never tackled — nobody knows the annual total.
• It is not solved by putting more people on the mailbox: it is solved by removing the latency between the message arriving and somebody able to decide knowing what to do.

How it fits the IRIS system

Connect in chaotic-sources mode — it reads a corporate mailbox, not people.

Connect in chaotic-sources mode.

The system returns to the logistics lead which racks to hold, what to reorder and who to warn. A proposed decision, not a notification that an email arrived. The human approves or corrects.

• Dedicated mailbox in silent copy and corporate messaging channel, live twenty-four hours a day.
• The model extracts intent — customer line stop, mix change, supply delay, delivery window change — and entities: schedule, variant, reference, hour.
• Connect cross-references against the active sequence and the planned expedition load.
• It returns the concrete action to the logistics lead: which racks to hold, what to reorder, who to warn. A proposed decision, not a notification.
• The human approves or corrects. Humans in command.

See the full IRIS architecture →

Before and after

Alert read when somebody can vs. alert processed instantly

Aspect	Current handling	With iLEAN Connect
Latency to action	1-3 h of human latency	Seconds
What reaches logistics	The forwarded email	The concrete action, cross-referenced with the sequence
Material in the wrong order	Found while preparing the rack	Avoided before building
Urgent transports	Absorbed as a diffuse cost	Reduced and measurable
Customer and supplier channel	Email and messaging, unchanged	Email and messaging, unchanged
Who decides	Whoever reads the email, when they read it	The lead, with the proposal in front of them

From one to three hours of human latency to seconds. Fewer urgent transports, fewer racks broken down, decisions taken with this minute's picture.

Impact estimate

Impact estimate for your plant — to validate against 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.

• Sequenced delivery with a reaction window measured in minutes and critical alerts arriving outside EDI.
• Pilot on the customer and transport mailbox, which is where what changes the day comes in.
• Indicative payback between 3 and 7 months depending on alert frequency and the average cost of special freight.
• What usually surprises first is the figure: almost no plant knows what it spent last year on unbudgeted urgent transport.

Estimated payback 3 to 7 months depending on the frequency of critical alerts and the average cost of special freight. *Estimate to validate*.

And the fair question from the production manager

“What if it misreads an email and reorders the sequence wrongly?” — it reorders nothing on its own: it proposes to the logistics lead, who decides. The AI task is <b>anchored</b> — extracting schedule, variant, reference and hour from a text — and there the best models drop below <b>1.5%</b> error ^{<a href="#source-openai">[1]</a>}. What changes is not who decides, but when they find out.

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