AI control of chilled ready meals — short shelf life, multi-SKU, zero margin for error.

The problem

Four variables at once, and no single system joining them.

The quality manager of a chilled ready meal plant looks at four dashboards on four different screens every shift:

1. Cold room and tunnel temperatures — in the SCADA, with records signed every morning on paper.
2. The packing machine's MAP parameters — gas mix, pressure, seal integrity, in a vertical PLC that almost nobody cross-checks with anything else.
3. Batch recipe and allergens — in the ERP, in the MES, or in an Excel sheet from the R&D department, updated each morning in a shared folder.
4. The printed expiry date — on the label, changed manually with every SKU launch and entrusted to one operator plus a checklist.

The classic system works 99% of the time. That 1% is what ends up in RASFF — and in chilled ready meals, where shelf life is measured in days, an alert is not managed with slack: it is managed by pulling the batch off the shelf and restocking within 24h, with all the reverse-logistics, destruction and brand-damage cost that implies.

How it fits the IRIS system

iLEAN does not add a fifth system — it seals the cracks between the four you already have.

The chilled ready meal problem is not a lack of information: it is information living in islands that, at the critical moment (the SKU changeover, the overnight cold room spike, the gas change on the packing machine), does not reach the decision-maker in time. iLEAN acts as the putty that fills those gaps, without asking you to change the SCADA, the MAP system, the ERP or the packing line.

Edge sees the label and the seal before shipping. Connect reads the cold room, the MAP, the recipe and the carrier's emails. The agent cross-checks and, if something does not add up, holds the batch. The person signs — never the other way around.

The three iLEAN pieces applied to chilled ready meals:

• Edge — a machine-vision (CNN) terminal over the tray sealer. It reads the printed label (date, batch, allergens), verifies the integrity of the MAP seal and triggers the actuator (light signal, ejector) if it does not match the expected recipe. It works without a network. If the plant loses WiFi, Edge keeps reading and holding.
• Connect — captures the batch recipe and the cold room temperatures whether they come from a modern ERP, the SCADA, a vertical MES or a forgotten Excel file. And it captures what arrives from outside: the carrier's 6:00 am email "the truck broke the cold chain for 90 min on the highway," the shift lead's WhatsApp "recipe change, SKU 412 goes vegan," at second zero, without anyone forwarding anything.
• Agent — cross-checks cold room temperature, MAP parameters, batch recipe, line history (what ran before, with what cleaning) and the image of the final package. If there is a deviation, it does not send an email at 10 pm: it holds the batch at shipping and alerts the quality manager through whatever channel they use. The person validates and signs; the shipment does not leave on its own.

See the full IRIS architecture →

Before and after

Manual control + checklist vs. cross-checked control with iLEAN

Aspect	Manual control + checklist	With iLEAN Edge + Connect + Agent
Overnight cold chain	Spike unseen, signed off next morning	Agent detects the spike and cross-checks affected batches
SKU changeover (allergens)	Paper checklist, shift sign-off	Edge verifies the label + agent verifies the cleaning
Batch MAP parameters	Trend in the PLC, never cross-checked with quality	Joined with recipe and label at second zero
Printed expiry date	Operator plus checklist	Edge vision reads it and compares with the batch recipe
Carrier's email	Mail folder, seen when someone looks	Connect escalates it before the next load
File for the IFS/BRC auditor	Rebuilt by hand, days	Automatic per-batch dossier, with photo and traces

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Chilled ready meal plant, 2-3 MAP lines, multi-SKU with daily recipe changes (vegan, gluten-free, lactose-free) and 7-14 day shelf life.
• Edge + Connect pilot on one line (camera over label and seal + integration with cold room, MAP and the batch recipe). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of documented incidents, the average recall cost at your retailer and the improvement in scrap from bad seals or misprinted expiry dates (reasonable estimates starting at ≥30% reduction).
• The hard lever is a single avoided recall plus the reduction of recurring scrap from bad MAP seals and undetected cold chain breaks.

And the operations director's reasonable doubt

"What if the agent stops a shipment by mistake and I run out of restock on the shelf?" — the agents have no hands on critical OT: they propose, the person decides and signs. And in anchored tasks (cross-checking a cold room reading with a recipe, comparing a temperature with a threshold), the best models brought error below 1.5% ^[1]. The third safety ring exists precisely for this: the hold is a proposal signed by the quality manager, not an autonomous decision.

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