The back pallet goes up three times: 18,164 levels we weren't counting

In a double-deep rack, some pallets live comfortably and some live trapped. The front one comes out easily: the forklift arrives, goes up, picks it and comes down. The back one doesn't. If the front slot is taken by another product, the operator first has to bring down the front pallet, then go up for the back one, and then go up again to return the front pallet to its place. Three maneuvers to pull a single pallet.

Anyone who has worked on the floor knows that. What we didn't know is that our own system was counting it wrong.

The error nobody could see

Our system measures how scattered each lot's pallets are: how many aisles they're in, how deep, at what height and how many maneuvers it takes to pull them. With that, the Planner builds re-slotting plans that state how much they save per day.

We were already counting the three maneuvers for the back pallet. But when calculating how much the mast goes up and down, the calculation made a single vertical trip: up to the level and back down. As if all three maneuvers happened at the same height without moving.

In reality, each maneuver goes up and down. A back pallet on the fourth level, with the front slot taken, doesn't travel 6 levels: it travels 18. The rule ended up this simple: levels = 2 × (level − 1) × maneuvers. Try it above with the level control.

It looks like a formula detail. It isn't.

What changed when we fixed it

We ran the new rule over all the class A inventory currently in rack, without writing anything, in memory only, to compare.

How much is that in time? With the speed assumptions we currently have approved (lifting with load at 0.3 m/s, lowering faster, 1.5 m between levels), each level costs on the order of 4 seconds. 18,164 levels come to about 20 forklift hours. At the $95 MXN an hour from the first installment, almost $1,900 every time those pallets ship. And they ship many times.

Let me be clear about what I know and what I don't: the double-deep forklift times still say “untimed” in our data. The exact number may move. What doesn't move is the direction: going up and down is the slowest and most delicate part of the move, and we were underestimating it by a third.

The good news (which is also a lesson)

The first thing we checked was whether the fix changed the priorities. If the lots the system calls “these are the worst” suddenly turned out to be different ones, everything planned before would be in doubt.

Result: zero lots changed position in the dispersion ranking or in the depth ranking. The order is set by horizontal meters and extra maneuvers, and those were already right. What changed was the size of the bill: the re-slotting plans we already made still point at the right pallets, but they save more than they said.

I like that for two reasons. First: separating “what's worse” from “how much it costs” protected us. An error in the cost didn't contaminate the priority. Second: a plan that paid for itself in ten days may pay for itself in seven. That changes what gets approved this week and what waits.

The auditor that found the wrong culprit

Every number the app shows is recalculated by an independent auditor, on its own, and the two have to match one by one. On October 1 they didn't match: the auditor said 629 lots with pallets in rack and the app said 367. A gap like that is scary.

When it said 629 versus 367, neither the app nor the auditor was wrong: the comparator was failing. Even the system that watches needs someone watching it.

When we investigated, neither the app nor the auditor was wrong. Both counted 629. The one failing was the comparator, the program that puts them side by side: it was subtracting from the app, a second time, the lots that have no pallet in rack (the ones in damages or in stage). It was fixed there, and the definition was written down with a real example so no one interprets it differently again.

After the fix: 6,871 figures match and 1 differs. That one is seven lots in the remnant count that the app counts and the auditor doesn't. We didn't hide it; it went into the plan by name.

The moral is uncomfortable: even the system that watches the system needs someone to watch it. And a definition that isn't written down with an example is a definition two programs will understand differently.

Why this matters beyond the rack

Vertical travel isn't just time. It's safety. Every maneuver at height is a moment with a load overhead and an operator looking up at the ceiling. A pallet that requires three maneuvers on the fourth level is three times that risk. Having the system count it right is what lets the Put-away agent stop proposing “back and high” for a product that ships every day.

And we're taking the same principle outside the warehouse: fleet, cold chain, SLA deliveries and replenishment. There are “back pallets” there too: the stop that forces you to unload the whole truck, the refrigerated product that gets opened three times before reaching its destination. If it isn't counted, it isn't seen.

Several of our numbers are still red. But this week one of them got more honest: we now know the back pallet goes up three times, and we're charging for it in the plan.

How much does the back pallet weigh in your operation? If you're counting it as a single lift, you're probably underestimating the cost of your double-deep rack. Tell me below.

Comments

comments · I answer every one. If you disagree, even better: say it with numbers.

• loading…

post comment