Cycle counting with bulk reads: designing a count that happens often

01 / FIELD NOTE

Keep the decision tied to the operating context.

The argument for cycle counting is usually made as a cost argument — smaller counts, less disruption — and that undersells it. The real advantage is diagnostic. A count taken close to the event that caused a discrepancy can still be traced back to it; a count taken six months later cannot, because too much has happened in between to isolate a cause.

Bulk reading is what makes the frequency possible. Counting is expensive per item when it is done one at a time, so the natural response is to do it rarely and comprehensively. When a pass can capture a whole shelf or pallet in the time it used to take to find and aim at one label, the unit of counting changes from the item to the area, and the rhythm can change with it.

Scope is the first design decision. A cycle count runs against a defined area, a defined category, or a defined movement frequency, and the choice determines what the count can detect. Counting by area finds misplaced stock and confirms what is physically present. Counting by category confirms a stock figure for a group that matters commercially. Counting by movement frequency concentrates effort on the items most likely to have drifted, which usually gives the largest return for a fixed amount of time.

Tag level sets the ceiling on what any count can report. If goods are tagged at pallet level, a bulk read confirms pallets and cannot confirm their contents; a carton missing from inside a read pallet is invisible. Before designing the count, decide which questions it must answer, and choose the tag level that can answer them. Discovering this mismatch after the labels are applied is expensive.

Bulk reads need a position, not a walk-by. Reading a shelf from the aisle captures tags on the shelf, tags on the far side of the racking, and tags on whatever is passing behind. A count read taken from a defined position, in a defined direction, at a defined power, is reproducible; one taken while walking varies with pace and angle and cannot be compared with the last one. The count’s value comes from comparing successive results, so reproducibility matters more than reach.

Coverage is worth measuring rather than assuming. A test at commissioning — placing known tags throughout the area and reading from the intended position — establishes which parts of the shelf are reliably read, which are marginal, and which are hidden. That map is the difference between a count whose gaps are known and one whose gaps are indistinguishable from missing stock.

The exception categories are the count’s actual output. Expected and read is the uninteresting case. Expected and not read suggests an item that moved without being recorded, an unreadable tag, or a read-position gap. Read and not expected suggests a receipt or transfer that was never posted, or stock belonging somewhere else. Read in the wrong place identifies misplacement directly, and read twice identifies a tag problem or an over-count. Each points somewhere different.

A count has to decide what to do about tags it hears but should not be counting. Stock staged for dispatch, goods already picked against an order, returns awaiting inspection — all of them are physically present and none of them belong in the count of available stock. If the read zone cannot exclude them, the count will report them as unexplained surplus, and the team will learn to disregard the exceptions.

Expected accuracy is not uniform, and treating it as though it were wastes effort. A bin of loose fasteners and a rack of serialized high-value units do not warrant the same counting frequency or the same tolerance for variance. Setting the rhythm by value, movement rate and consequence gives the same effort a better return than counting everything on the same schedule.

The result is only useful if it is comparable. That means the same area, the same read position, the same power settings and the same tag population, so that a change between one count and the next means something changed on the floor rather than in the method. A configuration baseline — what was counted, from where, at what settings — belongs with the count record, not in someone’s memory.

Reconciling is a workflow, not a report. Someone has to be able to see the exceptions, investigate the ones worth investigating, and record what was found, and that has to be easier than working around the system. A cycle count that produces an exception queue nobody clears is a cost with no benefit, and the team will revert to counting the way it did before.

Finally, the count should be judged on whether it changes anything. A count whose discrepancies are adjusted to match the record has converted a detection into a concealment. The point of counting often is that the cause can still be found — which only happens if the count is treated as evidence about the operation rather than as an exercise in making two numbers agree.