Product QR code with AI — the door to the batch digital record, not to a static page.

The problem

A QR with no batch behind it is a marketing trick — and the regulator has noticed.

Anyone can stick a QR code on their product and link it to a page of static PDFs. The problem starts when the QR has to mean something:

1. The end customer scans and wants to see their batch, not the generic datasheet. That batch data lives in the MES, in the ERP, in a spreadsheet on the R&D manager's laptop, on a LIMS control sheet — scattered.
2. The regulator requires the DPP by product family. Home appliances, batteries, textiles, construction materials — the Digital Product Passport is arriving, and the QR (or Data Matrix) is its carrier. It is not a differentiation option: in some sectors it is already compliance.
3. The printer is still carrying the previous batch QR. It is the most common and most expensive incident: SKU changeover, the printer was not updated, the whole pallet ships with the wrong QR. When someone scans and sees a product that is not theirs, it reaches quality and the retailer — and sometimes a recall.

The classic setup (a QR linked to a static web page) keeps working — until the customer, the retailer or the auditor demand the truth of the batch behind it. And the truth of the batch lives scattered.

How it fits the IRIS system

iLEAN links the QR to reality — without you throwing away your MES or your ERP.

The challenge of the product QR is not printing it: it is linking it to the truth of your plant in real time. iLEAN does not ask you to change the printer or the ERP; it acts as the filler that seals the crack between the scattered data (MES, ERP, LIMS, spreadsheets) and the endpoint that serves the QR to the customer.

Edge reads the printed QR. Connect joins recipe, batch and certificates. The agent builds the DPP. The person signs what goes to market.

The three iLEAN pieces applied to the product QR:

• Edge — a terminal with machine vision (CNN) over the marking or packaging line. It reads and decodes the printed QR, cross-checks it against the active MES order and verifies that it points to the correct batch. If there is a deviation, it holds the item before packaging and fires the actuator. It works without a network.
• Connect — captures the batch data wherever it lives: recipe from the ERP, parameters from the SCADA, controls from the LIMS, the batch photo taken by the shift supervisor, the supplier certificate that arrived by email. That is the precondition for the QR to mean something real.
• Agent — lives in Central. It recontextualises that data into the Digital Product Passport or GS1 Digital Link format, publishes it to the endpoint that serves the QR, manages differentiated permissions (what the end customer sees vs. what the B2B retailer sees vs. what the auditor sees), and assembles the compliance dossier (EUDR, CBAM, REACH/SCIP where applicable). The person validates the publication — nothing reaches the market without a human signature.

See the full IRIS architecture →

Before and after

Static QR vs. QR linked to the batch with iLEAN

Aspect	QR to a generic web page	With iLEAN Edge + Connect + Agent
Link to the real batch	Generic, by SKU	Specific to the batch and serial produced
Verification at print time	None	Edge reads the QR and verifies against MES on the line
Origin of the content	Static PDFs someone updates	ERP/MES/LIMS/Connect cross-checked by the agent
DPP / GS1 Digital Link compliance	Partial or absent	Native, compatible structure
Differentiated permissions	n/a — all or nothing	End customer / B2B retailer / auditor
Detection of the previous batch QR	Arrives as a complaint	Held on the line before packaging

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 lay it out so that management and compliance have an order of magnitude; we refine it during the diagnostic.

• Manufacturer with 1–3 marking/packaging lines, exporting to the EU/UK/USA, in a sector with the DPP on a close regulatory horizon (home appliances, batteries, textiles, construction materials).
• Edge pilot on one line + Connect for batch capture + DPP agent. First value expected within a few weeks: product leaves the pilot with a QR linked to the batch and an endpoint the customer and the retailer can scan.
• Indicative payback between 4 and 9 months, conservative. Levers: reduction of incidents from a wrong QR ≥30% (estimate), access to retailers that already demand a real-batch QR, savings in compliance audit preparation, and reduction of counterfeiting.
• The hard lever: a single recall avoided because of a wrong label or QR pays for the pilot. A single blocked listing at a major retailer over a non-compliant QR is avoided with the same system.

And the reasonable doubt from the CAIO / compliance

“What if the agent publishes wrong data in the DPP?” — hallucination is a problem of free generation, not of tasks anchored to the source. Recontextualising ERP/MES data into the DPP format is exactly that: reformatting, not inventing. On anchored tasks the best models brought the error rate below 1.5% ^[1]. And even so: nothing reaches the market without a human signature — the three safety rings require it by design. It fits the manufacturer's liability and the direction the AI Act is taking.

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