Warehouse slots: the address decides the pick

Where an item lives decides three things at once: how long the pick takes, how long the count takes, and how often the box goes back to the wrong place. The layout drawing matters less than the address rules behind it — a plain warehouse with strict addresses beats a clever drawing whose addresses change, because the drawing is read once and the address is read every time somebody touches stock. The rule flips at a site picking whole pallets from the door: there the walk is short enough that the address is a formality.

What makes a shelf into an address?

Four properties, and a code missing any one of them sends a picker to the wrong place with confidence.

It is unique — one code, one place, no "the one by the window". It is stable — an address is retired, never recycled, so a pallet under an old label is never mistaken for stock in a new position. It is readable in walk order — zone, aisle, rack, level, position. And it is printed at the place it names, at eye height, because an address that lives only in the system is one half the staff cannot use.

Then the system has to be the authority. The moment a slot code on a shelf has no record behind it, the receiver invents one, and the invented ones are the ones that never quite match. And an address is retired, never reassigned: a suffix that cannot be mistaken for a live code, or a new version of the whole map, is the same demand for identifiers that survive their context that the W3C makes of published data.

Should the fast movers sit by the door?

Yes, and the decision is worth almost nothing unless it is repeated. Velocity slotting — putting the items that appear on most order lines within the shortest walk — is not a drawing, it is a recurring job, because demand moves with the season, with a new supplier, with a product line that quietly dies. A map drawn once puts last year's bestseller at the door and this year's in the far corner, and nobody notices until the picking hours climb.

Rank by order lines, not by units. An item sold in pallets to three customers a month moves volume and almost no picking.

What is the difference between a pick face and a reserve?

The pick face is what a person can reach without a ladder or a forklift; the reserve is the bulk behind or above it. The split only works when replenishment is a task with a name attached, and when the system knows that reserve stock is not pickable.

Without that line, one of two things happens. Pickers climb, which is how a warehouse injures somebody. Or the system counts reserve units as available, promises an order the pick face cannot fill, and the short pick is found at despatch.

Decision	What it buys	What it charges
Pick face separate from reserve	A pick that never needs a ladder	A replenishment task with an owner
Fast movers near despatch	A shorter walk on most lines	Re-slotting that never really ends
One item per slot	An address that means exactly one thing	Space you cannot fill with anything else
Slot size set by the item	Fewer crushed and hidden units	A rack you cannot reconfigure cheaply

Why does a shared slot cause so many errors?

Because overflow has to go somewhere, and "nearby" is where it goes. The assigned slot is full, so the receiver puts the pallet in the next slot, on the floor at the end of the aisle, or on the level above — and the address in the system becomes a partial truth. The picker arrives and finds nothing, or finds a different batch, and the next count turns the mistake into an adjustment nobody can explain.

Two things fix it, neither of them discipline. Give the slot a capacity in the system, so an over-full put-away is refused and an overflow address is offered instead of improvised. And keep one item per slot: two products in one address make every count of that address an argument about which product the number belongs to.

How should a cycle count be routed?

By address, in walk order. A count list sorted by item name sends the counter back and forth across the building, so a short count takes an afternoon and the back corner gets skipped. A list sorted by slot is one pass, and it checks two things at once: the quantity, and whether the box is where the system says it is.

Frequency follows movement — fast movers often, slow ones rarely — but every address is counted at least once in the cycle. The shelf nobody has looked at for a year is where the model has drifted furthest.

The order to do this in

1. Walk the warehouse and write down the addresses that exist, including the pallets on the floor that have none.
2. Choose one format — zone, aisle, rack, level, position — at a fixed length, and print it everywhere.
3. Split the pick face from the reserve, and write down what a full slot means in units.
4. Put a capacity on every slot, and create the overflow addresses the system may offer.
5. Rank the items by order lines, place the fast movers within reach, and schedule the review, not just the first pass.
6. Label every rack edge before the first scan; a slot without a label is not yet a slot.
7. Switch counting to walk order, and count the address alongside the quantity.
8. Retire old codes deliberately — a suffix or a map version — and never hand one to a new position.

Where this goes wrong

• The code describes the contents. "Detergent-3" becomes wrong the day detergent moves, and then two addresses mean detergent.
• Aisle numbering starts at the door on one side and at the back on the other. Everyone who knows the building is fine; nobody new is.
• Overflow lives wherever there is space. The stock exists and the system cannot say where.
• Two sources of location. The system says one thing, a supervisor's spreadsheet another, and the picker trusts the spreadsheet.
• Labels printed small, at knee height. The address is correct and unusable.

What this does not cover

This is about the address and the slot, not the building: racking strength, aisle width, fire egress and forklift routes belong to an engineer, and the layout that satisfies them may not minimise the walk. It does not cover automated storage or robotics. It assumes enough space that a choice exists — in one room with four shelves only the address rule is worth arguing about. And none of it repairs a stock model that disagrees with itself: a good address makes a correct model fast and a broken one fast too.

We build the slot model into the system: one address format, a capacity per slot, counts routed in walk order, and a map that can be redrawn without renaming the shelves.