Raw milk traceability from farm to plant — one record for every liter received, not five islands nobody cross-checks.

The problem

Five different sources, none of them talking to the others.

Raw milk traceability is the textbook example of a process that has all the information — and has it in different places that do not talk to each other:

• The carrier's routing software — it knows which farms, in what order, how many liters, at what time. It lives in their system, or simply in their head and on the delivery note.
• The tanker's temperature record — on a chart recorder in the truck, sometimes only on paper, sometimes on an IoT device belonging to the carrier.
• The receiving report — what the dock operator writes down when the truck arrives: plate number, time, tanker, liters, first temperature reading.
• The intake analyzer — fat, protein, somatic cells, antibiotics. It prints on a strip or shows on an old panel, and somebody types it (or doesn't) into the system.
• The ERP/MES — where “the summary” gets dumped, by hand, late, and with the occasional error.

Five islands. When a food safety alert arrives, or a question from the regulatory council, or a suspicious pattern at one farm, somebody sits down to rebuild it by hand — cross-checking five sources that share no common format. It works, but at the cost of person-hours and of nuance slipping through. And nuance is what matters.

How it fits the IRIS system

iLEAN does not make you throw anything away — it listens on every channel and unifies.

The lack of integration across the farm-to-plant chain was not an oversight by the industry — it was a real problem with no good solution for decades. Connecting 5 different sources, without forcing the carrier to change tools and without scrapping the old analyzer, was simply not viable. That has changed. AI is, at last, the answer to that problem. iLEAN acts as the putty that fills the cracks between the 5 systems, without asking you to change any of them.

Connect listens on email, WhatsApp, phone calls, paper and old panels. Edge reads the analyzer. The agent unifies everything by tanker and by farm, at second zero. The person signs — a decision about a supplying farm is not taken by an agent alone.

The three iLEAN pieces applied to raw milk traceability:

• Connect — the plant's ear and voice towards transport. It captures email, WhatsApp, transcribed calls, a photographed paper report, the old chart recorder panel. And it captures inwards too: the dock operator dictates “tanker 27, probe reading 4 °C”, and it lands linked to the batch at second zero.
• Edge — reads whichever intake analyzer you already have (Bentley, MilkoScan, FOSS or any other), or reads it with vision if the output is a printed strip. It links the result to the tanker and the batch with no manual typing. It works with no network: if the dock loses WiFi, it keeps recording on the panel's own power.
• Agent — unifies the 5 systems by tanker and by farm, prepares the dossier for audits and the food safety authority, tracks patterns per farm so no signal is lost, and warns before a bad pattern becomes entrenched. The person signs — the decision about a farm is taken by quality, not by an agent.

See the full IRIS architecture →

Before and after

Five islands rebuilt by hand vs. one unified dossier per tanker

Aspect	Classic system (5 sources)	With iLEAN Connect + Edge + Agent
Route data	In the carrier's software or on the delivery note	Captured at second zero by Connect
Transport temperature	Chart recorder, sometimes on paper only	Integrated, linked to the batch automatically
Receiving report	Written out by hand by the operator	Voice/photo + integration → straight into the dossier
Intake analysis	Printout + manual typing afterwards	Read directly or by vision, linked to the batch
Dossier for the authority or auditor	Rebuilt by hand, hours or days	Generated by the agent, ready at close of day
Problem pattern at a farm	Discovered in the monthly report	Detected by the agent at the third occurrence

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.

• Mid-sized dairy plant taking raw milk from a multi-farm cooperative, several tankers a day, an existing intake analyzer, and a receiving report still partly on paper.
• First value expected within a few weeks: per-tanker dossier running, capture of the carrier's channels (email/WhatsApp/calls), integration with the analyzer.
• Indicative payback between 4 and 9 months, depending on how much administrative time goes into manual reconstruction, how often supplying farms cause problems, and the cost of an inspection carried out with weak traceability.
• Hard levers: office person-hours cut by ≥ 30% (a conservative estimate); early detection of a pattern at a farm; immediate defensibility in front of the food safety authority or the regulatory council. A single well-handled inspection pays for the pilot.

And the operations director's reasonable doubt

“What if the AI decides to block a farm without my approval?” — let us pre-empt that: it does not. A decision about a supplying farm is one of those that, in iLEAN, live inside the human validation ring: the agent prepares the proposal, presents the pattern with the data, and the quality manager signs or does not sign. The agent does not release the batch, does not block the supplier, does not write to the regulatory council on its own. In anchored tasks (cross-checking data, reformatting, recontextualizing), the best models bring error below 1.5% ^[1]. And even so, the sensitive decision is signed by the person — the three safety rings exist precisely for this.

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