Cycle counting: a practical guide

What cycle counting is and why it beats an annual count

A cycle count is a count of a small subset of inventory, such as a group of bins or a list of SKUs, done on a schedule throughout the year. Over a full cycle, every location or every item gets counted at least once, and the important ones get counted more often.

The annual physical count has two problems. It stops the warehouse for a day or more, and it finds errors months after they happened, when nobody remembers the mis-pick or the unrecorded damage. A cycle count that finds a variance in bin A-01-02 on Tuesday can usually be traced to Monday's transactions. That turns counting from an accounting ritual into a way to fix the process that caused the error.

Cycle counting methods

Most programs combine two or three of these. Pick the mix that matches where your errors come from.

• ABC counting: rank SKUs by value or pick frequency, then count the A items (the small share that drives most value or movement) most often, B items less often and C items least. It puts counting effort where an error costs most.
• Location-based counting: count every bin in an aisle or zone, in walk order, whatever it contains. It finds mis-slotted stock, items in the wrong bin and bins that should be empty, which SKU-based counts miss. It also gives you a clean way to prove every location was counted during the year.
• Opportunity counting: count when the work already brings someone to the bin. Typical triggers are a bin reaching zero, a quantity dropping below a low threshold, a short pick, or a failed putaway.
• Control-group counting: count the same small set of SKUs repeatedly for a few weeks when starting out. Repeated errors on a known set expose process problems quickly.
• Random sampling: count a random selection of locations to get an unbiased accuracy estimate alongside the targeted counts.

Blind counts vs informed counts

In an informed count, the counter sees the expected quantity. In a blind count, they see only the location and, sometimes, the item, and record what is physically there.

Informed counts are faster, but they invite counting to the number. A counter who expects 48 and sees roughly that many will often confirm 48 without counting the last layer. Blind counts remove that bias. They take a little longer and produce more first-pass variances, but those variances are real. Use blind counts as the default and handle speed with good labels and clear units of measure instead.

Be explicit about the unit being counted. A variance of 11 on a SKU stored in cases of 12 is usually a counter who counted cases instead of eaches, not missing stock.

How often to cycle count: building a schedule

Start from the number of counts you need per year, then turn it into a daily target your team can keep. A common starting point is counting A items monthly, B items quarterly and C items twice a year. Adjust from there based on the variances you find.

For example, a warehouse with 2,000 SKUs might classify 200 as A, 600 as B and 1,200 as C. That gives 200 × 12 + 600 × 4 + 1,200 × 2 = 7,200 counts a year. Over 250 working days, that is about 29 counts a day, which one person can usually finish in the first part of a shift.

Schedule counts when stock is quiet, such as before picking starts or after the last carrier pickup. Keep the daily list the same size so counting becomes routine rather than something skipped on busy days. If counts keep getting skipped, the schedule is too ambitious; shrink it and protect it. Opportunity counts sit on top of the schedule and do not replace it.

Hard freeze vs soft freeze

A count is only meaningful if you know what quantity it should be compared against. The classic answer is a hard freeze: block all movement in the location until the count is posted. It is simple and accurate, but it stops picking and putaway in that area, which is hard to justify every day.

A soft freeze keeps the warehouse working. The system takes a snapshot of the expected quantity when the count starts, then adds or subtracts any movements recorded in that location before the count is posted. New orders are steered away from the bin while it is being counted, but a pick already in progress can still finish.

For example, bin B-02-03 holds 50 units of brake pads BRK-220 when the count starts. A picker confirms a pick of 6 during the count, and the counter records 44. The expected quantity is 50 minus 6, which is 44, so there is no variance. Without that adjustment, the count would show 6 units of false shrinkage.

A hard freeze still has its place. Use it for a full physical inventory, for high-value cages where every unit matters, or when your system cannot compare against snapshot plus movements. For daily cycle counts, a soft freeze is almost always the better trade.

Handling variances: recount, investigate, approve

A variance is a question, not an answer. Work it through the same steps every time, and never correct a balance by overwriting the number.

• Recount: send a first-pass variance to a second count, by a different counter when possible. Many variances disappear here.
• Check open transactions: putaways not yet confirmed, picks confirmed late, receipts still at the dock, and returns not yet processed.
• Check neighbors: look in the adjacent bins and the same SKU's other locations for mis-slotted stock.
• Check units of measure: a case counted as one each, or a pack of six counted as six, explains many large variances.
• Approve by threshold: set a tolerance, for example up to 2 units and below a set value per line approved by the team lead, with anything larger going to a manager.
• Post with a reason: record the adjustment as an inventory movement with a reason code (count variance, damage found, mis-slot), so the history shows what changed and why.

How to measure inventory accuracy

Inventory accuracy is not one number, and the one you pick changes the story. Location accuracy is the share of counted locations where the physical count matched the system, within tolerance. For example, if you counted 500 locations and 485 matched, location accuracy is 97%. SKU accuracy does the same by item across all its locations.

Unit variance tells you the size of the errors. Always report absolute variance alongside net variance. If one bin is 20 over and another is 20 under, net variance is zero while absolute variance is 40, and something is clearly wrong. Value accuracy weights each variance by cost, which is what finance cares about.

Categorize every approved variance by cause: receiving error, putaway to the wrong bin, mis-pick, unit-of-measure confusion, unrecorded damage, or theft. After a few weeks the categories show where to spend effort. If most variances trace to putaway, better counting will not fix them; a mandatory location scan at putaway will.

Set the tolerance before you start measuring, and track accuracy weekly alongside your other warehouse KPIs. The trend matters more than any single week.

Starting a cycle counting program: the first month

The first month sets habits, so keep it small and strict. The goal is not coverage yet; it is a process that produces counts you trust and variances you can explain.

Expect the first weeks to surface more variances than you would like. That is the program working. Each variance you trace to a cause, such as a receiving step that skips the location scan or a pack size entered wrong in the catalog, prevents dozens of future errors.

• Week 1: fix the basics. Every bin has a readable label, every SKU has a barcode and a clear unit of measure, and counters know the difference between a case and an each.
• Week 2: run a control group. Count the same 20 to 50 SKUs every day and trace every variance to its cause.
• Week 3: add location-based counts of your fastest pick zone, in walk order, blind.
• Week 4: set tolerances and approval rules, classify SKUs into A, B and C, and publish the daily schedule.
• From then on: review accuracy weekly, and fix one root cause at a time rather than recounting the same bins.

Common cycle counting mistakes

Most weak counting programs fail for the same handful of reasons.

• Showing expected quantities to counters, which turns counts into confirmations.
• Posting first-pass variances without a recount.
• Counting SKUs but never whole locations, so mis-slotted stock is never found.
• Counting while stock moves, without a freeze or a snapshot-plus-movements comparison.
• Adjusting balances directly instead of posting movements with reasons.
• Measuring net variance only, which hides offsetting errors.
• Treating counting as a fix instead of tracing each variance to the process that caused it.

How NextStock runs cycle counts

NextStock runs cycle counts as blind, location-first counts on the floor app, with count lists in walk sequence. A bin under count is soft-frozen: allocation skips it and picking warns but can proceed. The expected quantity is the snapshot at task start plus the net ledger movement since, so a concurrent pick does not show up as shrinkage.

A first-pass variance creates a recount, assigned to a different counter when one is available. Only a variance that survives the recount goes to a manager for approval, and approved variances post as movements with reasons in the inventory ledger. Count sheets are available when you want to count on paper, and count accuracy is one of the KPI reports.