GS1 DataMatrix with AI — one mis-encoded Application Identifier stops the importer, not the printer.

The problem

A readable DataMatrix is not a correct DataMatrix.

The verifier on the case packer says "OK". The American customer rejects the shipment two weeks later because their DSCSA system cannot parse AI 17 (expiry date) — the date was right, but the FNC1 separator was missing. The GS1 serialization failure patterns we see over and over:

1. Badly built AI/FNC1 structure — the code prints, the code reads, but it does not parse against the GS1 standard; the repository spits it out.
2. Borderline ISO 15415 grade — it passes today with the printer's new camera, it fails tomorrow with the scanner on the Rotterdam dock.
3. Broken parent-child aggregation — the master case declares 12 packs and holds 11 plus one duplicate; nobody notices until landing.
4. EPCIS events reconstructed afterwards — commissioning was uploaded by someone with a spreadsheet yesterday, because the aggregation system never sent it.

The classic verifier checks whether the code can be read. iLEAN checks whether the code makes sense — for the regulator in the destination country, for the wholesaler receiving it, for the EPCIS repository it is about to be published to. The failure either stays in the plant or it stays nowhere.

How it fits the IRIS system

iLEAN does not add another case packer — it adds the brain the case packer does not have.

The problem is not printing: it is that the verification chain between printer, case packer, palletiser and repository is broken at the joints. iLEAN acts as the sealant that closes those joints, without ripping out your serializer or your aggregation system.

Edge watches every step (print → unit → case → pallet) with GS1 criteria. Connect moves EPCIS to the retailer hub. An agent holds parent-child coherence together and signs the events at the moment they happen. The person validates.

The three iLEAN pieces applied to the GS1 DataMatrix:

• Edge — a terminal with computer vision (CNN) that decodes the DataMatrix, splits the AIs by FNC1, validates the GS1 structure, measures ISO/IEC 15415 quality and checks the GTIN check digit. It blocks the unit if the grade or the structure does not comply. It works with no network, on the light available at the panel.
• Connect — emits and consumes EPCIS 2.0 (JSON/JSON-LD) or EPCIS 1.2 (XML) depending on the retailer, wholesaler or regulator hub. It also captures the importer's email changing a quality threshold, the messaging thread with the adhesive supplier, the spreadsheet kept by the serialization manager.
• Agent — holds parent-child across three levels (pack → case → pallet) in real time; orchestrates the EPCIS events (commissioning, packing, shipping) and proposes corrections when a case comes up short. The person signs; the system does not decide alone.

See the full IRIS architecture →

Before and after

Classic verification vs. GS1 DataMatrix with iLEAN

Aspect	Classic verifier	With iLEAN Edge + Connect + Agent
What gets validated?	That the code is readable	GS1 AI/FNC1 structure + check digit + ISO 15415 grade
SKU / regulatory profile change	Manual reconfiguration of the verifier	Profile per family (DSCSA/FMD/cosmetics…) applied automatically, validated by a person
Print quality	Pass / fail	A-F grade + print head and laser trend
Parent-child aggregation	Counts tags	Vision of the real units + agent reconciliation
EPCIS events	Uploaded afterwards, from a spreadsheet	Signed and emitted at the moment of the event
Importer rejection	Arrives two weeks later	Impossible — the rejection is avoided at the outbound dock

Impact estimate

Impact estimate for your plant — to be validated with your own numbers.

The block below is an estimate to be validated with the concrete data of your line. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Packaging line with unit-level plus master-case serialization (pharma, cosmetics, premium food), exporting into 2-3 regulatory regimes.
• Edge pilot on the case packer with a GS1 profile plus EPCIS integration from Connect. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, dominated by: a reduction of ≥ 30% in rejections caused by bad printing or mis-encoded AIs plus a drop in returns caused by broken aggregation.
• Hard lever: the cost of one single return from a US or EU port under DSCSA or FMD pays for the pilot.

The demanding standard

In sectors such as automotive, the quality bar sits around the order of 25 PPM (parts per million). In pharma under DSCSA or FMD the threshold is not measured in PPM at all, it is zero rejections: no invalid serial can enter the chain. What moves the needle is not a better verifier — it is verifying against the standard that actually applies, inline, before aggregation. That is exactly the kind of anchored task where AI is reliable: it checks against a defined structure instead of guessing ^[1]. The Edge does it with GS1 criteria, not the OEM reader shipped with the printer.

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

See also: UDI medical device with AI → · Export track & trace with AI →