Product hologram with AI — the seal properly applied on every unit, at eight in the evening too.

The problem

Four ways the seal fails, one single consequence.

The security hologram fails in four ways, all of them expensive:

1. Missing — the reel shortfall left a gap on one unit. It reaches the shelf with no seal, the consumer notices, trust drops.
2. Out of position — shifted, lifted, folded by poor feeding on the labeller. It looks like a counterfeit to the customs inspector.
3. Damaged — scratched in internal transport, faded by temperature, badly die-cut. It looks like a fake by the time it reaches the customer.
4. Generic, reused — the hologram from another reference fitted by mistake during the SKU changeover. It passes human visual control: it "looks right", it is simply the wrong one.

The classic system (operator at end of line, magnifier, checklist) works 99% of the time. And its quality drops after the sixth hour, like every human inspection. That 1% is the one that reaches the shelf or the customs inspector — and the conversation about batch recall, return to plant and brand damage reopens.

How it fits the IRIS system

iLEAN does not add one more station — it puts a trained eye and memory on the one you already have.

The hologram problem is not a lack of security technology: it is that the human eye fades at the end of the shift and the decision depends on one person at the control station. iLEAN acts as the filler that closes that gap, without asking you to change the labeller, the hologram supplier or the packaging line.

Edge sees the hologram under multi-angle lighting, on every unit. Connect reads the serial and the batch. The agent cross-references with the ERP and holds if something does not add up. The person signs — never the other way round.

The three iLEAN pieces applied to hologram verification:

• Edge — terminal with machine vision (CNN) mounted at the labelling station or at final control, with controlled multi-angle lighting. It reads the optical pattern of the hologram — laser, hot-stamping, kinegram, micro-typography, UV — and fires the actuator (ejector, light tower, hold) in milliseconds when it is missing, badly placed, damaged or not the one for that SKU. It works with no network: inference runs on the terminal, it does not depend on the WiFi.
• Connect — captures the printed or marked serial code (datamatrix, QR, OCR), the batch and the production plan from the MES. When one reference carries a reinforced seal and another the standard one, Connect brings the active rule to the station from second zero.
• Agent — cross-references the hologram reading, the serial, the batch and the active SKU rule. If everything adds up, it lets the unit through and records it. If not, it holds the unit, alerts the quality manager and prepares the dossier (hologram photo, serial, batch, expected rule). The person validates and signs; the line does not restart on its own.

See the full IRIS architecture →

Before and after

Human inspector at end of line vs. multi-angle vision with iLEAN

Aspect	Operator + magnifier + checklist	With iLEAN Edge + Connect + Agent
Coverage	Sampling, or 100% with a drop-off at the end of the shift	100% constant, no fatigue drop
Detection of a missing hologram	Good at the start, irregular at the end	Constant 24/7
Hologram from another SKU after a changeover error	"Looks right": it passes	Cross-references the active SKU from the MES: holds
Damaged by poor feeding	Detected only if badly torn	Detected by deviation from the optical pattern
File for an auditor or the brand owner	Shift sheet, photo if someone took one	Automatic dossier with photo, serial, rule
Operation with no network	n/a	Edge keeps operating on cabinet power

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 product. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.

• Packaging line with a hologram on the neck, label or cap (spirits, premium cosmetics, tobacco, OTC pharma, lubricants, OEM spare parts).
• Edge pilot on one line: camera with multi-angle lighting + actuator + MES integration for the active SKU rule. First value expected within a few weeks: first run with 100% verification.
• Indicative payback between 4 and 9 months, depending on the documented cost of past recalls and pressure from the distributor / customs.
• Reduction in units with a defective hologram reaching the shelf of ≥ 30% expected against your current baseline. The hard lever is a single recall avoided — one recall pays for the pilot.

And the quality manager's reasonable doubt

"What if the camera gets it wrong and throws false positives that stop the line?" — hallucination is a problem of free generation, not of anchored tasks. On tasks where the AI only compares an image against your own pattern trained on your authentic units, the best models brought the error below 1.5% ^[1]. And even so, the hold does not release on its own: the person signs the restart. The confidence threshold is configurable per plant (auto-approval or mandatory human validation); the three safety rings are there for precisely this.

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