Dairy cow — the collar knows, the meter knows, the ERP knows — and nobody knows it together.

The problem

Four apps per cow, none of them talking to the others, and the heat is missed.

A modern dairy cow does not suffer from a lack of technology — she suffers because her reality lives in four silos that do not talk to each other:

1. The GPS collar records activity, rumination, heat and lying time in its own app.
2. The milking parlour, conventional or robotic, records kilos per milking, conductivity, somatic cells and milking time in its own system.
3. The vet writes down treatments, calvings, dry-offs and diagnoses on a sheet, in a different app, or straight into the farmer's messaging app.
4. The farmer's or the co-op's ERP collects litres delivered, price per kg, quality analyses and the settlement.

The farmer knows that cow number 247 is off. The collar said so yesterday; the milking dropped this morning; the vet has a note in another app. But nobody has the individual record of 247 stitched together first thing today. The classic system (apps + eye) works most of the time. The few times it does not, it turns into missed heats, late mastitis and chronic lameness — and every one of those has a hard cost.

How it fits the IRIS system

iLEAN Tracer is not a fifth app — it seals the cracks between the four you already have.

The problem on a modern farm is not a lack of data: the collar, the parlour, the vet's app and the ERP do have the information. The problem is that it lives on four heterogeneous islands, mostly without a common API, and the individual record per animal exists in no system — only in the farmer's head. Tracer is the filler that fills those gaps, without asking you to change the collar, the parlour, the vet's app or the ERP.

Edge verifies the animal in the parlour. Connect captures the four apps. The agent stitches the individual record per animal and per milking. The person — farmer and vet — still decides.

The three iLEAN pieces applied to the dairy cow:

• Edge — a terminal with vision and a reader in the parlour that verifies that the animal entering the stall matches the one expected from the collar and from the programmed queue. If a cow slips in out of order, it records it and warns the farmer. It works with no network — on a rural farm connectivity comes and goes, and Edge keeps operating on the power from the panel.
• Connect — captures the collar (through the vendor API or the standard integration), the electronic milk meter, the vet's app and the ERP. It also captures what comes in from outside — an email from the external vet with an analysis result, a message from the co-op with a quality figure — without anyone forwarding anything.
• Agent — stitches the individual record by animal number and by milking. It flags deviations that would otherwise be found late (conductivity rising + rumination falling = early mastitis; high activity + predicted heat = insemination window). And it generates the digital farm logbook without anyone retyping it.

See the full IRIS architecture →

Before and after

Four apps per cow vs. an individual record stitched with iLEAN Tracer

Aspect	Collar + parlour + vet app + ERP	With iLEAN Tracer
Individual record per animal	Only in the farmer's head	Stitched per milking, on a single screen
Yield per milking	Only in the parlour	Cross-checked with heat, health and ration
Mastitis detection	When the cell count shows up days later	When conductivity and rumination diverge
Heat detection	Collar app, no milking cross-check	Collar + yield cross-checked per animal
Digital farm logbook	Retyped by hand from 4 apps	Generated automatically, signed
Change of parlour or collar make	History broken	Tracer stitches both sides, no swap needed

Impact estimate

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

The block below is an estimate to be validated with the actual data from your milkings and your animals. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.

• Dairy farm with 150-500 cows in milk, conventional or robotic parlour, modern collars already installed, an active vet app.
• Tracer pilot on one parlour + the collar + the ERP (integration with the parlour make and the collar make you already run, no vendor swap). First value expected within a few weeks: the individual record stitched together and heats detected by crossing collar + milking.
• Indicative payback between 4 and 9 months, depending on the number of cows in milk, the annual mastitis incidence and the current heat detection rate.
• The hard lever is early detection of mastitis and heat: one cow pregnant on time or one mastitis stopped within 24h outweighs the whole investment. Expected reduction of indexable incidents ≥ 30% once the pilot is stabilised.

And the farmer's fair objection

“What if the AI gets it wrong and flags a mastitis that is not there?” — hallucination is a problem of free generation, not of anchored tasks. On tasks where the AI merely cross-checks the conductivity from the milk meter with the rumination from the collar and puts the pattern in front of you, the best models brought the error below 1.5%^[1]. And even then, nothing critical is decided alone: Tracer puts the case in front of the vet, who signs. The three rings are there precisely for this.

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