Beef carcass identification with AI — the animal cannot get lost between the ear tag and the hook.

The problem

An identity that is lost at stunning is worth exactly as much as no identity at all.

European regulation is clear: every bovine animal carries two ear tags, a bovine passport and an entry in the national system. The plant has an intake record, a slaughter order and a destination for every carcass. But the chain breaks at three classic points:

1. Entry to slaughter. The tag is read by hand, or the order in the pen is simply assumed. One change of order for animal welfare reasons and the next reading no longer matches.
2. Stunning to hook. The animal goes onto the hook and, from that instant, there is no ear tag any more — there is a hook tag that someone had to assign correctly. One crossed assignment and two carcasses swap identities forever.
3. Shift or day changeover. The operator who knew the last assignment goes home, the next one opens a notebook and starts from the best guess available.

When the inspector later asks for a dossier on a health alert, or the customer asks for traceability of one specific cut back to the farm of origin, the answer is "we'll rebuild it" — and rebuilding is never the same as having.

How it fits the IRIS system

iLEAN does not add another registry — it seals the crack between the ones you already have.

The problem with bovine identification is not a lack of records, it is that the records live on islands and the chain breaks at the physical moment (intake, stunning, change of hook). iLEAN acts as the putty that seals those cracks with vision + assisted capture, without asking you to change SITRAN, REGA, your ERP or the line.

Edge reads the ear tag with vision and links it to the hook at second zero. Connect cross-references the bovine passport and the national registries. The agent delivers the per-carcass dossier before the inspector asks for it. The person signs whenever there is doubt.

The iLEAN pieces applied to beef carcass identification:

• Vision + Edge — a CNN camera at the stunning / slaughter point reading both ear tags. Natural redundancy: if one tag is dirty or damaged, the other one usually is not; if both fail, the animal goes to the manual procedure with all the evidence captured. Edge signs off the assignment to the hook in milliseconds. It works without a network — the identity is stored locally and synced afterwards.
• Connect — captures the digital bovine passport (or a photo of the paper one where it is not digital yet), links it to the farm code, and publishes to your ERP/MES. Integration with SITRAN/REGA goes through an API where that is allowed and through assisted capture where the administration imposes a manual channel. If a health alert arrives by email at 10 p.m. ("batch X from farm Y is placed on hold"), it enters the system at second zero and the agents locate every affected carcass still in the plant without waiting for Monday's meeting.
• Agent — keeps the per-carcass dossier live and available: image of the ear tag as read, associated bovine passport, farm code, hook position, cutting route, destination of every piece. When the inspector asks for traceability on a specific cut, the dossier already exists — it is not rebuilt.

See the full IRIS architecture →

Before and after

Ear tag read by hand + notebook vs. linked identity with iLEAN

Aspect	Manual reading + notebook / Excel	With iLEAN Vision + Connect + Agent
Ear tag reading	Operator at intake	CNN vision, both tags, redundant
Assignment to the hook	Order is assumed	Linked at second zero, with a photo
Shift changeover	The classic failure point	Identity inherited, never rewritten
Integration with SITRAN/REGA	Manual re-entry	API or assisted capture, no re-keying
Health alert from the farm	Email at night, meeting on Monday	Affected carcasses located in seconds
Dossier for the inspector	Rebuilt by hand, weeks	Already built — photo of the tag included

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 beef slaughterhouse, one main slaughter line, current reconciliation incident rate per working day known.
• Vision + Edge pilot at the stunning point + integration with the bovine passport and the national registry. First value expected within a few weeks: identity closed on the line, per-carcass dossier available.
• Reduction in tag-to-carcass reconciliation incidents of ≥ 30% in the first months, and a sharp drop in the time it takes to put together a dossier for the inspector. Conservative estimate; the ceiling depends on how much of today's reconciliation work was avoidable versus caused physically (a torn-off tag).
• Indicative payback between 4 and 9 months, depending on volume and on the cost of each documented incident. The hard lever is the combination of office time saved + fast response to health alerts + the ability to serve markets that demand full farm-level traceability without a day's delay.

And the quality manager's reasonable doubt

"What if the vision system misreads a tag and assigns the wrong identity?" — hallucination is a problem of free generation, not of anchored tasks such as reading an ear tag code and comparing it against a record. In anchored tasks, the best models brought error below 1.5%^[1]. And even so, what is critical is never decided alone: when a reading is not sharp, the system flags the animal for the manual procedure and the person signs. iLEAN's three safety rings guarantee that an identity is never assigned without human validation when there is reasonable doubt.

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