Mike Churkin

Finding a given item was almost impossible; people were burning hours hunting for stock. It looked like an efficiency problem, but it wasn’t — it was a customer one. Every order we couldn’t fill on time was a sale walking out the door.

So I started with data, before any racking or robots: we rebuilt the item master, so the warehouse actually knew what it held, then sorted the packaging product by product. Get that layer wrong and everything downstream inherits the mess. From there the sizing — how many locations, on shelves and in racking, for the demand ahead — then the processes, written out, and the people they’d need.

We ran the supplier tender and stayed through the build. I didn’t leave until the new warehouse was running.

I walked it and mapped the processes. Pick slots were being topped up carton by carton — the operator climbing down to move boxes by hand, then lifting the pallet back into the racking.

My suggestion was simple: replenish with full pallets instead. The objection was fair — rotation, swapping stock around the pick face, slow and costly. So we settled it with numbers. A process model built on standard times put the full-pallet method at three times more efficient, even counting the extra return flow.

They put it in and handled more volume than before — without buying a single extra reach truck.

We mapped the processes together, wrote the requirements, and picked a modest system with just enough in it — then put two small AI models on top.

The first placed stock by where demand was heading: fast movers near the dispatch doors, slow ones deep in the back — sharper than the usual ABC analysis. The second handed pickers their orders so they walked less, grouping the picks to cut wasted steps — sometimes pulling three orders out of one aisle in a single pass.

Picking productivity doubled. No drawn-out rollout, no clever algorithm that goes stale the week the flow shifts.