Cycle counting: setting a cadence that earns its hours

You block out a Saturday, count the whole warehouse, and find the shelf disagrees with the system on roughly a fifth of your SKUs. Six weeks later it disagrees again, by about the same amount. The full count told you the number was wrong. It did not tell you why, and it cost you a Saturday to learn it.

Cycle counting replaces that Saturday with a short count on a normal working day. This page covers what cycle counting is, how to set the cadence from value at risk rather than from the calendar, which SKUs earn the most frequent count, and when a full count is still the right call.

What is cycle counting?

The meaning is literal: cycle counting is counting a small subset of your SKUs on a rolling schedule, daily or weekly, rather than everything at once. Across a defined period every item gets counted at least once. The value is not the count itself: it is catching a variance while its cause is still close enough to find.

A quarterly full count gives you one accurate day and eighty-nine uncertain ones. You correct the number and learn nothing about what produced it.

How often should you cycle count?

Set frequency by value at risk per counting hour, not by the calendar. Value at risk is what one unit costs multiplied by how fast an error compounds: how quickly the item sells, and how close it sits to the end of its selling window. High figures earn a weekly count. Slow, cheap, long-dated stock earns a quarterly one.

Most published advice sets a cadence by ABC class, ranked on annual sales value. That ranking was built for durable goods, and it misses the thing that decides cost in food and drink: dating. A cheap SKU with six weeks of shelf life left can cost more per counting hour than an expensive one with a year, because an error on the short-dated item turns into a write-off before the next count would have found it.

Now follow one mistake. A case of your fastest line is miscounted at goods-in in January. Under a single annual count it sits there until December: for eleven months the system promises stock you do not have, and you find out when an order fails. Under the ladder above it is an A-tier product, so the gap shows up in February, while the paperwork is still in the folder and the person who booked it in remembers the pallet.

Note what did not change. The C tier is counted no more often than before, because cheap, long-dated stock gains nothing from a faster cycle. You will count more across the year in total, and always on ordinary days with the floor still running.

Which SKUs earn the most frequent count?

The ones where an error costs the most before anyone would otherwise notice. That means fast sellers, expensive units, anything with a short selling window, and anything a person handles often: repacks, split cases, and items stored in more than one place. Movement and handling decide the tier, not where the item sits.

Handling is the part people leave out. A pallet that arrives, sits, and ships whole is barely exposed. A case that gets split three ways for three channels passes through four decisions, and every one of them is a chance to write down the wrong figure. Count what people touch.

Two figures feed the tiering directly. One is how much buffer the selling window absorbs, which tells you how quickly a short-dated item turns from stock into a write-off. The other is what actually left the distributor’s warehouse, because a SKU that looks slow on your side can be moving fast one step down the chain.

When does a full physical count still win?

A full count earns its shutdown when you need one dated figure for everybody at once: a year-end valuation, a warehouse move, or a system migration. It answers what the number is today. It does not answer why the number keeps going wrong. Run it for the event that needs it, then go back to cycling.

The two are not rivals. Cycling keeps errors small enough that the full count stops being frightening, because there is no longer a quarter of accumulated error waiting inside it.

What makes a cycle counting program stick?

A count that records a variance and stops there teaches nobody anything. Every variance needs a cause written beside it, from a short fixed list, and someone reading which cause repeats. The cadence keeps the number honest. The cause list is what makes the number stop moving in the first place.

Three habits carry it: a fixed slot at the same time each day, the same short list of causes every time so the pattern is countable, and a quarterly re-tier as products speed up or slow down. Skip the cause column and you have built a very disciplined way of rewriting the same wrong number forever.