Traceability is born at the dock, or it is not born at all

The problem

There are only two ways out, and both are bad.

Chemical raw-material receipt is where the traceability of everything the plant makes afterwards is born. And it is the point under the most time pressure: carrier waiting, dock occupied, and one person acting as a human scanner. There are only two bad ways out. Either the truck waits forty-five minutes charging demurrage, or it gets typed in a hurry and the lots end up badly linked, which surfaces months later in a quality investigation that can no longer be closed. The certificate of analysis, meanwhile, ends up in a folder of PDFs nobody ever cross-references with the lot.

• Chemical raw material receiving is where the traceability of everything the plant later makes is born, and it is the point under most time pressure.
• Either the truck waits forty-five minutes racking up demurrage, or it gets typed in a hurry and the lots end up badly linked.
• The second path does not raise an error at the time: it surfaces months later, in a quality investigation that can no longer be closed.
• The certificate of analysis, meanwhile, ends up in a folder of PDFs nobody ever cross-references with the lot it certifies.

How it fits the IRIS system

Connect in photo mode — without asking the supplier to change its delivery note.

Connect in photo-of-the-analogue mode. Vision AI recognises the delivery note's layout without asking the supplier to change theirs, and extracts the tabular structure. From the pallet labels it reads GS1-128, QR, barcode or plain text by OCR. From the certificate of analysis it extracts the analytical parameters and ties them automatically to the raw-material lot. Validation with one tap on the dock's industrial tablet, then posted to the ERP through the API.

What really changes is not the speed: it is that the certificate of analysis stops being a loose PDF and becomes tied to the raw material lot and, transitively, to every product batch that used it.

See the full IRIS architecture →

Before and after

Today's receiving versus captured receiving

Aspect	Today	With iLEAN Connect
Truck held at the dock	45 minutes	Under 10
Lot linking	Wrong or late	At the moment of unloading
Certificate of analysis	A loose PDF in a folder	Tied to the lot and to the products that used it
Pallet labels	Typed	GS1-128, QR, barcode or plain text
Delivery note format	One per supplier	Recognized without asking for changes
Receiving and warehouse licenses	Two or three	One, the supervisor who validates

Impact estimate

Estimated impact — to validate with your own numbers.

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.

• 💰 CFO angle: the receiving and warehouse module typically requires two or three active licenses. With Connect only the supervisor validates with a tap, without entering the ERP: typical 50-70 % reduction in recurring licenses.
• From 45 minutes of truck held at the dock to under 10, truck after truck.
• Raw material traceability linked at unloading, not rebuilt months later.
• The certificate of analysis goes from dead file to evidence cross-referenceable with every product batch.

CFO angle — direct ERP license saving: the goods-in and warehouse module typically requires two or three active licences (goods-in supervisor, backup, supervisor). With Connect only the supervisor validates with one tap without entering the ERP. Typical reduction of 50-70% in goods-in/warehouse module licences, plus avoiding the bulk-capture module many ERPs sell separately, plus the direct saving on carrier demurrage. Estimated payback of 3 to 8 months, plus the recurring annual saving. *Estimate to validate*.

And the fair question from the production manager

«Does it read a dirty or creased label at the dock?» — most pallet labels carry standard symbology, which is direct reading; flat optical recognition only comes in when the label is printed without it. In that case it is an anchored task and the best models drop below 1.5% error [1]. And above all there is the cross-check: if label and delivery note disagree, the discrepancy is highlighted rather than resolved silently.

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