Walk the Floor: Count It Now or Chase It in November

Every building I ran held a 98.5% inventory accuracy standard.

Not 95%, which gets cited as world-class. Not the industry average, which sits somewhere around 83%. Ours was 98.5%, measured a specific way: total locations checked against the locations where a quantity had to be changed. Count a thousand locations, find fifteen wrong, you're at standard. Sixteen and you're not.

The definition matters, because accuracy gets measured a dozen ways and most of them are generous. Ours was location-driven and unforgiving. A location was either right or it wasn't. No partial credit for close.

Inventory accuracy and shrink are two different questions

Alongside accuracy we tracked something separate — the total adjustments to the on-hand value balance, plus and minus. That was shrink.

They are not the same number and they don't answer the same question. Accuracy tells you whether the location a picker is about to travel to is correct: right SKU, right count. Shrink tells you what the errors cost. A building can look clean on one and bleed on the other.

Most buildings track one of them well and the other barely. Usually it's shrink, because finance asks for shrink. Nobody outside the four walls asks for location accuracy — which is exactly why it drifts.

In an e-commerce facility carrying thousands of SKUs, location accuracy is the one that decides your day. Every pick location has to hold the correct SKU and the correct count, because there's no one standing there to catch it. The system says go here and take three. If the location has two, or has the wrong item, the error is already downstream before anybody knows.

The rule that mattered most: don't stop at the location

Here's the discipline that did the most work.

We set flags on large adjustments. When a count turned up a variance big enough to trip the flag, we didn't correct that location and move on. We counted the entire SKU — every location it lived in, whether or not those locations were in the count plan that day.

The reasoning is simple. A large variance in one location is rarely a one-location problem. It's a receiving error, a putaway error, a mis-slot, a replenishment that went somewhere it shouldn't have — and whatever caused it almost certainly touched the other locations holding that SKU. Correct the one you happened to count and you've found part of it. The rest stays in the building. Three of us compared notes on exactly this in the first episode of Question of the Week.

The variance you found is not the variance you have.

Why this is a July conversation

Counting an entire SKU across every location is a real interruption. You're pulling people off production to chase something that might turn out to be nothing.

In July you can afford that. Volume is soft, the building has slack, and an afternoon spent on a full-SKU count costs you close to nothing.

In November it isn't a decision you get to make. The building is running flat out, every body is committed, and the honest answer to "should we count the whole SKU" becomes "we'll look at it after peak." So you fix the location, you move on, and the rest of the error stays in the building — hiding inside your busiest and most expensive weeks.

The standard guidance is to count outside high-traffic periods. That isn't really advice about scheduling. It's an admission that during peak, counting stops.

Which means the accuracy you carry into November is the accuracy you built in July and August. There is no catching up once the volume arrives.

Walk your pick locations now. Set the flags now. Run the full-SKU counts now, while a two-hour interruption is only a two-hour interruption.

Count it now, or chase it in November.