The capital pain of any winery, solved in 4 rings

The problem

A wrong vintage on a bottle costs more than the wine inside it.

a batch/vintage/origin incident is the biggest reputational and certification risk in the wine sub-sector.

• A batch, vintage or origin incident is the biggest reputational and certification risk a winery faces, far above any single defect on the line.
• It rarely comes from one big failure. It comes from small ones in a row: a batch defined right, a label typed wrong, a bottle nobody read, a sign-off with no evidence.
• Each control on its own looks fine, and the bottle still leaves with the wrong vintage on it.
• When it is found, it is found at the distributor or the importer, and what is at stake is the certification and a relationship built over years.

How it fits the IRIS system

Four coordinated rings — all the previous cases, checking each other.

4 coordinated rings —

No single ring is new: the ERP, the labeler, the camera and the sign-off already exist in some form. What is new is that they check each other, and that the line stops the moment two of them disagree, before the case is closed. That is the difference between a control system and a coordinated one: the discrepancy is visible in the cross-check even when each ring, looked at alone, reports that everything is fine.

• the ERP defines batch and vintage in the active order,
• the labeler prints what the ERP publishes,
• the Edge camera reads what's printed on the bottle and matches it against the order,
• the evidence pack ties the previous three together with the winemaker's sign-off. JIDOKA AI halts the line on any mismatch; SMED AI keeps batch changeovers fast.

See the full IRIS architecture →

Before and after

The four rings, and the incident each one stops

Ring	What it does	What it stops
1 · ERP	Defines batch and vintage in the active order	Bottling without a clear reference
2 · Labeler	Prints exactly what the ERP publishes	Typing errors on vintage and back-label
3 · Edge camera	Reads the printed bottle and matches it to the order	A label that differs from the order
4 · Evidence pack	Ties the three together with the winemaker's sign-off	A batch released with no proof
Line stop by JIDOKA AI	Halts bottling on any mismatch	The case leaving with the discrepancy
Fast changeover by SMED AI	Keeps batch changeovers short	Control that slows bottling down

latent incident risk every campaign → discrepancy caught right on the line, before the case leaves the winery.

Impact estimate

Impact estimate — to be validated with the winemaker and general management.

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.

• Estimated payback 6-12 months, measured against the first incident avoided. One incident avoided is usually enough to cover the whole system.
• Protection against a certification incident, the kind that can put a denomination of origin at stake.
• Protection of the multi-year distributor and importer relationship, which is harder to rebuild than any batch. Importers increasingly ask for that proof before renewing.
• And latent incident risk every campaign becomes a discrepancy caught on the line, before the case leaves the winery.

protection against a certification incident and protection of the multi-year distributor/importer relationship. Estimated payback of 6-12 months against the first incident avoided. Estimate to validate with the winemaker and general management.

And the fair question from the production manager

"Won't the line stop every hour for nothing?" — JIDOKA AI stops on a verifiable mismatch between sources, not on a hunch. Reading batch and vintage from a printed label and comparing it with the order is an anchored task, where the best models drop below 1.5% error [1]. When the doubt comes from the model rather than from the data, it goes to the winemaker instead of stopping the line.

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