Counting a little, often, instead of everything once a year — and why it finds causes rather than just differences.
Cycle counting is the practice of counting a small subset of stock continuously, so that every item is counted several times a year without stopping the warehouse. High-value or fast-moving items are counted more often than slow ones. The point is not the count itself: it is that a discrepancy found within days can still be traced to its cause, while one found at year-end cannot.
The annual count tells you that you are wrong, not why. By the time a difference surfaces in December, the transaction that caused it happened in March and nobody can reconstruct it. Cycle counting narrows the window to days, which turns stock accuracy from an accounting exercise into an operational one that can actually be improved.
Usually by ABC analysis — value, movement, or both. The classification determines count frequency.
A items perhaps monthly, B quarterly, C once or twice a year. The target is that every item is counted at least annually and the valuable ones far more.
The system issues a short daily list, by location or by item. Short lists get done; long ones get deferred.
The counter should not see the expected quantity. A displayed figure is a strong hint and compromises the count.
The adjustment is the easy part. The value comes from asking why, and fixing the process that caused it.
A warehouse holds 4,200 SKUs. Classified by annual movement value, about 340 are A items, 980 are B and the remaining 2,880 are C. The team has one person who can give roughly ninety minutes a day to counting.
Counting A monthly, B quarterly and C annually produces 340 × 12 plus 980 × 4 plus 2,880 × 1, which is 10,880 counts a year, or about 44 per working day. At ninety minutes that is roughly two minutes per count — tight but achievable for bin-level counts of discrete items, and clearly not achievable if each count means weighing or opening cases.
The useful part of this arithmetic is that it is done before the schedule is published. A cycle counting programme that is quietly impossible gets abandoned in the third month and taken as proof that cycle counting does not work. If the number comes out impossible, the honest responses are to widen the C interval, narrow what counts as an A item, or ask for more time — not to publish the schedule and hope.
| Term | What it means |
|---|---|
| ABC classification | Ranking items by value or movement so counting effort follows importance rather than being spread evenly. |
| Count frequency | How often a class is counted — A items perhaps monthly, C items once or twice a year. |
| Tolerance | The variance threshold below which a count is accepted without investigation. Setting it too tight buries the team in noise. |
| Reason code | The recorded cause of a variance. Without it you know the size of your problem and nothing about its source. |
| Blind count | Counting without seeing the expected quantity, so the counter cannot unconsciously confirm it. |
If every discrepancy is simply posted away, accuracy never improves and the count becomes a monthly write-off ritual.
Counting accessible locations and skipping the awkward ones produces a reassuring and meaningless accuracy figure.
Blind counting is the whole discipline. Once the number is visible, the count tends to confirm it.
Cycle counting decays quickly without someone reviewing accuracy trends and acting on causes.
| Metric | What it tells you | What distorts it |
|---|---|---|
| Count accuracy by class | The share of counts within tolerance, split A/B/C. One number across all classes hides where the problem is. | Including zero-stock locations, which count as accurate and inflate the figure. |
| Variance by reason code | Which process is causing error — receiving, picking, a shared pick face, unrecorded moves. | A large 'unknown' bucket, which means the codes are not being used honestly. |
| Counts completed vs scheduled | Whether the programme is actually running or quietly slipping. | Rescheduling missed counts forward, which makes completion look perfect. |
| Net vs absolute variance | Absolute variance is the real error; net variance can be near zero while two large errors cancel. | Reporting net only, which is the single most common way accuracy gets overstated. |
We build and support the systems this page describes. Start with a free working proof of concept — no cost, no commitment.