Every trade show for the last five years has featured machine-vision pallet grading. It is the obvious target: grading is repetitive, labour-intensive and apparently rule-based. It is also the task in this yard least suited to automation, and the enthusiasm for it is a good illustration of how automation decisions get made backwards.
Why grading resists it
A camera can measure. It can tell you a board is cracked, a fastener protrudes, a footprint is 47½ inches. What it struggles with is the judgement that sits on top: whether this particular hairline check matters given the lane this platform is going to, whether that staining is surface mould or decay, whether the pattern of damage across this whole trailer load is worth telephoning a customer about.
Our graders average nine seconds per platform and make about seven judgements in that time, several of which depend on context no camera has. We tried a conveyor-based line once. It was slower than three people and an eye, and we sold it within the year.
Where automation would genuinely pay, in order
- Nail removal. Dangerous, slow, repetitive, zero judgement required, and the single biggest bottleneck in the remanufacturing stream. Dismantling machines exist and are under-adopted.
- Inventory counting. Knowing what is in a four-acre yard in real time is a genuinely unsolved problem that costs everyone money, and it needs no judgement at all.
- Backhaul routing. Currently a person looking at a map every morning. It is a constraint-optimisation problem and software is better at it than we are.
- Moisture monitoring in storage. Sensors are trivially cheap and nobody uses them, so mould gets discovered rather than predicted.
- Treatment batch documentation. Already partly automated and should be entirely — the chamber knows its own curve; tying it to a shipment reference should not involve retyping.
The pattern
Every item on that list is unglamorous, involves no computer vision, and would not look good in a trade show booth. Which is roughly a complete explanation of why the industry is investing in automated grading instead.
The wider version of this mistake
It is not specific to pallets. Automation effort tends to flow toward the most visible labour rather than the most automatable labour, because the visible labour is what a manager walking the floor notices. The grader with a clipboard is conspicuous. The hour a day somebody spends reconciling treatment logs is invisible and far easier to eliminate.
What we are actually doing about it
Passport numbers burned into stringers, so traceability does not depend on anyone writing anything down. Chamber logs tied to shipment references automatically. A routing review every morning that is slowly becoming a tool rather than a habit. And three trained people on the sort line, which we expect to still be the arrangement in a decade.
Measurement versus judgement
The distinction that organises this whole question. A camera is excellent at measurement: dimensions, presence or absence, surface features, colour. It is poor at judgement, which is the act of weighing a measurement against a context that is not in the frame.
Grading is 20% measurement and 80% judgement. Is this hairline check at the notch going to matter for a platform heading to a plant-internal lane with light loads? Is that discolouration blue stain or early decay? Does the pattern across this whole trailer load suggest a customer's dock problem worth a phone call? Every one of those requires information the camera cannot see.
The nail-removal case, costed
Dismantling is the genuine bottleneck in our remanufacturing stream. It is slow, repetitive, physically punishing, involves zero judgement, and is the single most common cause of minor injury in a pallet yard. It is also the step that gates how much short-board material reaches the millwork bench — currently about a third of what is generated, because the bench is limited by de-nailed supply rather than by demand.
Dismantling machines exist, they are not new, and they are under-adopted across the industry. They are also not photogenic and would not draw a crowd at a trade show, which is approximately the full explanation.
Why visible labour attracts automation effort
A manager walking the floor sees three people grading. They are standing in the open, doing something repetitive, and they are conspicuous. The hour a day somebody spends retyping chamber temperatures into a spreadsheet is invisible, sits at a desk, and is trivially automatable.
This is not a pallet-industry pathology. It is a general bias toward automating what is salient rather than what is tractable, and it is why so many automation programmes deliver less than their business case while an obvious unglamorous win sits untouched two doors down.
What we have actually automated, and what it cost
- Passport numbers burned into stringers. Traceability no longer depends on anyone writing anything down. A branding tool and a database.
- Chamber logs tied to shipment references. The chamber already recorded its own curve; the work was connecting it to a lot number instead of retyping it.
- Routing review against two lists each morning. Still a person, slowly becoming a tool. This is where the next real gain is.
- Nothing on the sort line, and we expect that to remain true. Three people and three eyes, trained in house.
The prediction we would defend
Automated grading will arrive, and it will arrive as an assistant rather than a replacement: a system that measures and flags, with a person making the call. That is a useful product and nobody will be able to sell it with a photograph of an empty sort line, which is why it is taking longer than it should.
Priya, operations · Pallets Recycle, 1113 Ingram Ave, Richmond VA · published June 16, 2026 · 936 words. Corrections welcome — tell us if we have this wrong.