The pallet gets checked before loading, not after

The problem

At loading pace, a damaged pallet looks like any other.

A torn wrap, ripped film, or a leaning pallet that could topple in transit are easy to miss under time pressure. It generates a complaint, a return, and if the client is in food or medical, it can add a quality incident on top of the logistics one.

• At the shipping dock several trucks load in parallel and the pace does not let up, especially in the last hour before the departure slots.
• Torn wrap, ripped film or a pallet leaning enough to topple in transit are easy to miss under time pressure, and occasional sampling only catches a few of them.
• The damaged pallet becomes a claim, a return and a replacement shipment, and nobody can prove whether it left the dock damaged or was damaged on the road.
• If the client is in food or medical, a logistics incident becomes a quality incident on top, with its own report and its own questions at the next audit.

How it fits the IRIS system

Edge — local vision on every pallet, milliseconds, before loading.

It infers in milliseconds per pallet with local CNN vision, with no need to send the image to the cloud. The model is trained on the pallet-handling defects typical of this kind of operation.

The point is not to replace the dock team's eye but to cover what it cannot at peak: every pallet, every truck, every shift. The camera never gets tired at the end of the afternoon, which is when the leaning pallet tends to go through. And every pallet leaves an image of its condition at the dock door, which settles the argument about where the damage happened. For a food or medical client, that image is also part of the evidence that the wrap and packaging were intact when it left the warehouse.

See the full IRIS architecture →

Before and after

Today's loading check versus Edge at the dock

Aspect	Today	With iLEAN Edge
Coverage	Occasional visual sampling	100% of loaded pallets
Torn wrap or ripped film	Seen if someone looks	Detected before loading
Leaning pallet	Topples in transit	Stopped at the dock door
Where the image is processed in the warehouse	—	Locally, no cloud
Proof of condition at loading	None	An image per pallet
Borderline case	Loaded anyway	Escalated to the shipping lead

Before: occasional visual sampling lets damaged pallets slip through to the client. After: 100% inspection of every pallet before it loads onto the truck.

Impact estimate

Impact estimate — to be validated with your 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.

• Estimated payback 5-12 months, depending on the volume shipped and last year's damage claims.
• Lower cost of claims and replacement shipments, which are usually paid twice: the goods and the second delivery.
• From occasional sampling to 100% inspection of everything that loads, on every dock and every shift.
• And an image per pallet that shows its condition when it left the dock, useful when a carrier or client disputes where the damage happened.

Payback 5-12 months · 100% inspection of every loaded pallet. It cuts the cost of claims and replacements, moving from occasional sampling to 100% inspection of everything that loads.

And the fair question from the production manager

“What if it stops good pallets and slows loading down?” — classifying wrap, film and lean on a pallet in a fixed position at the dock door is an anchored task, where the best models drop below 1.5% error [1], and the model is trained on the pallet defects of this kind of operation. A doubtful pallet is not silently rejected: it is escalated to the shipping lead, who decides in seconds. And every decision the shipping lead makes on a doubtful pallet feeds back into the model, so borderline cases become fewer over time.

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