Inventory says it is there. The shelf says otherwise: a cycle-count routine

The screen shows six units. The picker finds two. Now an order is waiting while someone checks the neighbouring bins, receiving area and packing bench.

Correcting the number helps that order move forward. Preventing the next shortage takes another step: finding out how the record drifted from the shelf.

A cycle count checks a defined portion of your inventory on a recurring schedule. The useful output is a verified quantity plus an explanation for discrepancies. A workable routine makes room for both.

Choose a count unit that exposes the problem

Count by SKU and location, adding lot, serial number or stock status where your products require it. “We have 100 units somewhere” does not help a picker sent to an empty bin.

Define the unit of measure before anyone starts. If the system holds individual units, a sealed case of 12 is 12 units only after the team verifies the case quantity and contents under your counting procedure. Mixed cartons need an actual count.

Include locations recorded as empty. An unexpected carton in an empty bin can explain a shortage elsewhere. Record unidentified items as exceptions rather than assigning them to the nearest-looking SKU.

Build a schedule around risk

Start with the items most likely to disrupt orders or hide a costly error: frequent picks, repeated shortages, high-value stock, confusing variants and items handled in several units of measure. Include newly introduced products while the team learns their packaging and labels.

Here is a starting cadence to adapt, not an industry standard:

Review the groups after a month. Increase attention where errors persist; check that the remaining stock still receives coverage. A programme that counts the same easy bins repeatedly can produce impressive activity totals while avoiding its hardest locations.

Control movements before taking the snapshot

For a simple manual routine, temporarily pause picks, replenishment, putaway and transfers for the selected locations. Finish or explicitly account for open tasks, record the cutoff time and tell the floor when those bins will reopen.

This is a local pause, sized to the work you can complete promptly. You do not need to stop the whole warehouse for a few bins.

Some systems support counting while transactions continue. Oracle's inventory documentation describes both closing locations and accounting for transactions during a cycle count. Use that approach only when your system and process reconcile movements against the count time. An undocumented pick halfway through a spreadsheet count creates a false shortage.

Count blind, then investigate differences

Give the counter the SKU and location without displaying the expected quantity. Oracle defines a blind count as one that excludes the system balance from the count sheet or device. The purpose in this routine is to have the person record what is physically present.

Record an explicit zero for an empty location. A blank entry should mean unfinished work, never a presumed match.

When the count differs, have a second person recount before changing stock. For a small team, require an independent review of the recount and supporting evidence. Then check the movement history: recent receipts, transfers, picked orders, cancelled orders, returns, damaged stock and adjustments.

Look at the locations that the transactions point to. A case placed in the wrong bin may require a location correction. It does not necessarily represent lost inventory.

Approve the correction and preserve the evidence

Set approval rules based on unit differences, inventory value and product risk. Record the original system quantity, first count, verified count, adjustment, reason and approver. Keep unresolved causes visible even after a justified quantity correction.

Do not let a financial tolerance hide a repeated operating error. A one-unit difference in an inexpensive SKU might need little approval effort, but ten similar incidents can expose a broken receiving or picking step.

Reopen the location only after the physical stock and system record agree under the completed count. Confirm that any held tasks can resume against the corrected quantity.

Keep accuracy and variance separate

Use explicit definitions so gains and losses cannot cancel each other on the dashboard.

Consider this hypothetical count of five SKU-location records:

The signed net variance is zero: minus five plus five. The absolute unit variance is ten: add the magnitudes of both differences. They describe different things.

For this routine, define exact-match record accuracy as records with no difference divided by records counted. Three of the five records matched, so accuracy is 60%. That percentage describes these five records, not the entire warehouse or the percentage of individual units that are correct. If you also report dollar variance, value each SKU's difference consistently and show gross discrepancies as well as the net adjustment.

Close one recurring cause each week

Group discrepancies by cause: wrong-bin putaway, unrecorded movement, unit-of-measure confusion, receiving error, picking error or unresolved. Give the most persistent cause an owner and a specific correction.

If two similar variants keep being mixed, inspect their labels and locations. If cases are repeatedly received as single units, fix the receiving conversion and train against a real example. Count those locations again after the change.

Protect a recurring time slot for this work. Count fewer locations well if necessary, while maintaining the coverage schedule. The routine earns its place when the same discrepancy stops coming back.