A retailer order change in a dairy processing plant — the planner’s inbox is where the waste is lost.

The problem

The retailer's order change is where the waste is lost, not on the line in steady state.

In a dairy processing plant — grated, slices or portions for private label — steady-state production is predictable: once the order is launched, the line cuts, weighs and packs without surprises. The problem appears before packing, when the chain's buyer changes the order and that change does not reach whoever plans in time:

• Critical changes by email or WhatsApp — the notice of a bring-forward, a cancellation or a volume change lands in one specific person's inbox or phone, and they cannot keep it open in real time between the line, other orders and the rest of the shift.
• Product packed without its optimal destination loses value every day it goes unplaced — and in grated and slices, with an especially short commercial window, that margin runs out in a matter of days.

A notice at 09:15 can break three orders planned for that same afternoon. If the planner does not read it until 11:30 — because they were on the line or in another meeting — the product has already been packed with no destination, and what was a simple order change becomes waste or an urgent sale at a low price. It is the inbox's natural bottleneck, not the factory's.

How it fits the IRIS system

Connect in chaotic-sources mode — it reads the retailer's email and WhatsApp before the window is lost.

The order change needs no integration project with the retailer or an ERP change. The message that already arrives by email or WhatsApp needs to be read, understood and crossed with the active orders the instant it enters — without waiting for the right person to have a moment. That is what Connect in chaotic-sources mode is for.

A message enters the inbox or WhatsApp Business. Connect reads it at second zero, extracts what the retailer asks and on which order, and returns a concrete action to the planner. Nothing is touched in the ERP without a person confirming it.

How Connect operates in chaotic-sources mode on the retailer's order change:

• The mailbox and WhatsApp Business silently copied 24/7 — Connect receives a copy of the mail and the WhatsApp Business where the orders and their changes arrive, replacing nobody and without changing the channel the retailer already uses.
• The LLM extracts intent and entities — a bring-forward, a cancellation or a volume change as the intent; SKU, customer, date and volume as entities. In seconds, the message stops being loose text and becomes structured data.
• Immediate cross-reference with the active orders — Connect compares the extracted intent against the open production plan and detects which specific orders are affected by the change.
• A concrete action returned to the planner — not a summary of the email, but an actionable proposal: "reschedule order X with the alternative SKU, cancel order Y, you gain Z units without a destination". The planner sees the option already built, they do not reconstruct it by reading the thread.
• Human confirmation before touching the ERP — Connect neither reschedules nor cancels anything on its own. The person confirms the action and only then does it move to the ERP or the planning system.

See the full IRIS architecture →

Before and after

An inbox with no cross-reference vs. an order change managed with Connect

Aspect	Inbox with no cross-reference	Order change with Connect
Detecting the order change	Email/WhatsApp read when the planner has a moment — 2-3 h of latency	Read and extracted at second zero, 24/7
Cross-reference with active orders	Manual, reviewing the production plan by hand	Automatic, an immediate cross-check with the open orders
Action on the plan	The planner decides without full impact visibility	A concrete action suggested — reschedule, cancel, an alternative
Product with no destination	Packed anyway, waste or an urgent sale per the commercial window	Reassigned before packing — the waste avoided
Confirmation before touching the ERP	Not applicable, the whole circuit is manual	Human confirmation required, nothing executes alone
The change's traceability	Reconstructing the email/WhatsApp thread after the fact	Message, extraction and action archived by order

Impact estimate

Impact estimate for your plant — to be validated 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.

• Dairy processing plant (grated, slices, portions) with private label orders from one or several retail chains.
• Connect pilot in chaotic-sources mode on the planning team's mailbox and WhatsApp Business — without touching the ERP except on human confirmation. First value expected within a few weeks.
• Indicative payback between 3 and 7 months, depending on the frequency of private label order changes your plant handles. Estimate to be validated.
• Expected reduction of product packed with no destination, with the biggest impact on grated and slices for their short commercial window. Estimate to be validated with your data.
• The hard lever is a batch reassigned in time: the same product that was going to waste or sell urgently finds a destination because the change was read at second zero.

And the fair question from the production manager

"What if Connect misinterprets a message and reschedules something it should not have?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI limits itself to recontextualizing a specific data point from one medium to another (reading a retailer's message and comparing it with the open orders), the best models brought the error below 1.5% ^[1]. And even then, nothing critical is decided alone: Connect proposes the action and the person confirms before touching the ERP. The three safety rings are there precisely for this.

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