About 61% less time on inbound cargo matching — Meitetsu World Transport
Tested Published
Programme: Aichi Digital Island Project "TECH MEETS" / Aichi Prefecture
What we measured
| Measurement | Result | Scope and conditions |
|---|---|---|
| Less time on inbound cargo matching | ~61% | Within the measured scope of inbound handling: 370 seconds down to 142.72 seconds Source: Presented at AICHI INNOVATION CHALLENGE 2026 on 18 March 2026 (proof of concept run 4 February 2026) |
Every figure is a measurement taken within the stated scope. It is not a guarantee that another site under other conditions would produce the same result. Read the primary source →
The situation
Meitetsu World Transport handles international freight. At goods-in, staff match the text data a shipper sends — item, quantity — against the cargo that actually arrives.
That check was done by eye. With many line items and similar-looking packaging, it took time, and swings in volume were absorbed by adding people.
What we built
We built a system that matches product images and item data registered by the shipper against photographs of the cargo taken in the warehouse.
A worker photographs the cargo; the system compares it with the registered product image and returns a match or a mismatch. The step that had been a person reading text and looking at a box became an image-to-image comparison.
What we measured
In the proof of concept run on 4 February 2026, the work went from 370 seconds to 142.72 seconds within the measured scope — about 61% less time.
That scope is the matching step within inbound handling; it excludes the transport and paperwork on either side. The figure was presented as a co-creation project result at AICHI INNOVATION CHALLENGE 2026, held by Aichi Prefecture on 18 March 2026.
About this project
This work ran as a co-creation project under the Aichi Digital Island Project “TECH MEETS”, organised by Aichi Prefecture. Design, delivery and reporting were all done with the prefecture and its programme operator alongside us — this is not a result we assessed on our own.
FAQ
Is that 61% off the whole inbound process?
No. We measured the matching step within inbound handling, not the transport or paperwork around it. In that scope the work went from 370 to 142.72 seconds. The figure for a whole process depends on where you draw the boundary.
Could we drop this into our own warehouse as-is?
Not as-is. The setup assumes the shipper has already registered product images and item data. Without that, the first question is what serves as ground truth. We would start by walking through your current matching procedure.
Do the cargo photographs leave our site?
Where the data sits is decided per project. If it includes material that cannot leave your network, we design for a closed setup. See [how to handle data that cannot go to the cloud](/en/media/cloud-ni-dasenai-data/).
What did the proof of concept expose?
Accuracy depends heavily on how current and how detailed the shipper's registered product images are. The further those images are from the actual packed state, the more cases go to a person. We recommend checking that precondition before discussing model accuracy.