Every multichannel system depends on one thing it cannot check by itself: whether the shelf matches the screen. If the system says 12 and the bin holds 9, every channel is selling three units that do not exist. A yearly stocktake finds that months too late. Cycle counting finds it within days, a few items at a time, without closing the warehouse.
This guide is written for small teams: one building, a few people, hundreds to a few thousand SKUs.
In short: split your products into A, B and C classes, count A items most often, add exception counts for anything that looks wrong, count blind, recount differences above a tolerance, fix the cause, and post every adjustment with a reason. For a few hundred SKUs, under half an hour a day is usually enough.
Why Cycle Count Instead of a Yearly Stocktake
- No shutdown. You count a small slice every day while orders keep shipping.
- Errors are found while they are fresh. A count a week after a wrong pick can be traced; a count eleven months later cannot.
- Effort goes where it matters. Your fastest and most valuable items get the most attention.
- Accuracy improves over time. Because each variance gets a cause, the same mistake stops repeating.
Counting also has an accounting side. US Treasury regulations say that book (perpetual) inventory records should be verified by physical inventories at reasonable intervals and adjusted to match. A documented cycle count program is one way to do that; ask your accountant whether you also need a full year end count.
Decide What to Count: ABC Classes
Rank your products by the value that moves through them in a year:
Annual value = Units sold per year x Unit cost
Sort from highest to lowest and split:
| Class | Typical Share of Annual Value | Typical Share of SKUs | Starting Count Frequency |
|---|---|---|---|
| A | About 80% | About 20% | Every 30 days |
| B | About 15% | About 30% | Every 60 days |
| C | About 5% | About 50% | Every 120 days |
The splits are a rule of thumb, not a law; use the breakpoints your data shows. Some teams rank by units picked instead of value, because pick activity is what creates errors. Either works, as long as you are consistent.
Add Exception Counts
Schedules catch slow drift. Exceptions catch problems now. Count an item outside its schedule when:
- a channel showed a different quantity than your system,
- a picker found the bin empty or short, or the packer scanned the wrong item,
- stock went negative or more is reserved than is on hand,
- an item shows zero on hand but keeps selling,
- a return, transfer or receipt for it had a problem,
- it is close to its reorder point (fewer units to count, and the number matters most right before you reorder).
Exception counts usually find more errors per minute than scheduled ones.
How Many Items to Count Each Day
Items per day = (A items x A counts per year + B items x B counts per year + C items x C counts per year) / Working days per year
Example. A warehouse with 400 SKUs has 80 A items counted 12 times a year, 120 B items counted 6 times and 200 C items counted 3 times:
(80 x 12) + (120 x 6) + (200 x 3) = 960 + 720 + 600 = 2,280 counts a year 2,280 / 250 working days = about 9 items a day
At two or three minutes per item, including walking and scanning, that is under half an hour a day for one person. Add a few exception counts and you have a routine that fits around shipping.
Try it on your own data: start free, no card needed.
How to Run a Count, Step by Step
- Print or open today's list of items and their bins.
- Handle the cut off. Count when no orders for those items are being picked, or record which orders were picked during the count.
- Count blind. The counter sees the item and the bin, not the system quantity, so the count is not nudged toward the expected number.
- Count every location the item lives in: the pick bin, overflow, returns, a staging area.
- Count in the selling unit. Convert cases and inner packs to single units before entering the number.
- Compare the count with the system quantity.
- Recount any difference above your tolerance, ideally by a different person.
- Investigate confirmed differences before adjusting: check recent receipts, picks, returns and transfers for the item.
- Post the adjustment with a reason code (receiving error, mis-pick, damage, theft, unit of measure, unknown), so it is recorded rather than silently overwritten.
- Fix the cause if one is found, and note it.
Tolerances and Recount Rules
A tolerance decides when a difference triggers a recount and investigation. Tighter for items that matter more:
| Class | Tolerance | Recount If |
|---|---|---|
| A | Zero units, or up to 1% on very high counts | Any difference |
| B | Up to 2% | Difference above 2%, or above a set value |
| C | Up to 5% | Difference above 5%, or above a set value |
| Any | Value limit | Difference worth more than an amount you set, such as $50 |
Small differences within tolerance are still adjusted; they just do not need a recount.
Measure Accuracy
Inventory record accuracy is the share of counts where the system was right within tolerance:
Record accuracy = Counts within tolerance / Total counts x 100
Track value too, both ways:
Net variance = Sum of (Difference x Unit cost) Gross variance = Sum of (Absolute difference x Unit cost)
Net variance can look small because a missing $200 and a surplus $190 cancel out. Gross variance shows how much the records were really off. Example: in a week of 50 counts, 46 were within tolerance (92% record accuracy), the net variance was minus $35 and the gross variance was $410. The headline loss is small; the record keeping problem is not.
Watch the trend week by week. A rising accuracy rate means the root cause fixes are working.
Find and Fix Root Causes
| Symptom | Common Cause | Fix |
|---|---|---|
| Surplus after a delivery | Receiving counted the purchase order, not the boxes | Count what arrived, scan at receiving |
| Shortage on one variant, surplus on another | Picker took the wrong size or color | Scan to verify at packing; separate look alike items |
| Steady small shortages | Damages, samples or giveaways not recorded | A quick way to record a write off with a reason |
| Surplus after returns | Returns restocked physically but not in the system | Process returns through the system the same day |
| Big differences in multiples | Cases counted as units, or the reverse | Define the selling unit and count in it |
| Components short, kits fine | Kits assembled without recording components | Record assemblies, or sell bundles from components |
| Stock in two places at once | Transfer recorded at one end only | Transfers that track stock in transit |
Before You Start: Set Up for Good Counts
A cycle count program is only as good as the warehouse it counts. A few things to put in place first:
- Give every unit an address. Label bins and shelves, and record which bin each product lives in. Counting "somewhere in aisle 3" wastes time and misses stock.
- Clear the backlog. Process unreceived deliveries, unprocessed returns and pending transfers before the first counts, or every count will show a difference that is really paperwork.
- Agree on reason codes for adjustments, and keep the list short enough that people use it.
- Decide who counts and who approves. The person who picks an item should not be the only person who counts and adjusts it.
- Start with your A items. A full count of the A class gives you a clean baseline for the products that matter most, and shows the most common errors quickly.
Counting While Orders Keep Shipping
The main risk of counting during the day is a pick or a receipt landing between the count and the comparison. Three simple habits handle it:
- Count each item at a quiet moment, such as before picking starts or after the last carrier collection.
- If you must count during picking, note the time of the count and check for picks of that item after it before adjusting.
- Never adjust an item that has an open receipt or transfer in progress; finish that first, then recount.
Common Mistakes
- Counting with the expected number visible. People find what they expect to find.
- Adjusting without recounting. A single miscount becomes a permanent error.
- Counting only the pick bin. Overflow and returns areas hold the missing units.
- Treating adjustments as the end of the job. Without a cause, the same error comes back next month.
- Letting the list slide on busy days. Skip a day if you must, but never a week; the gap is where drift grows.
A Weekly Routine for a Small Team
- Every day: count the scheduled list plus any exception items, ideally at a quiet time.
- Every day: recount differences above tolerance before adjusting.
- Every Friday: review the week's adjustments by reason code and value.
- Every Friday: pick one root cause and fix it.
- Every month: refresh the ABC classes from the latest sales.
- Every quarter: check the record accuracy trend and adjust tolerances or frequencies.
Tools That Make It Faster
- Bins with labels, so every unit has an address and counts go by location.
- Scanning with a phone camera or a cheap Bluetooth scanner, so the right item is counted.
- Blind count screens that hide the expected quantity.
- A ledger that records each adjustment with who, when and why.
- Pack station scanning, which prevents many of the errors you would otherwise find in counts.
How Invechar Runs Cycle Counts
Invechar proposes a short cycle count every day, such as "Count 12 SKUs at Main Warehouse today", with the items most likely to be wrong first: a channel that disagreed about the stock in the last 14 days, a wrong item or extra unit scanned at the Pack Station, odd stock such as more reserved than on hand, and an ABC cadence that counts top sellers every 30 days, B items every 60 and C items every 120. Counts can be blind, run from a phone camera or a scanner, show system, counted and change side by side for review, and post every variance to the stock ledger as a correction, so each change is traceable. Bins, pick waves and the Pack Station stop many errors before they happen.
Read about stock counts and cycle counts, bins and locations, the Pack Station and scanning, or see why counts matter for preventing overselling.
Sources
- Electronic Code of Federal Regulations, 26 CFR 1.471-2, Valuation of inventories, on verifying book inventories by physical inventories at reasonable intervals. Accessed October 8, 2026.
- ABC classification, counting frequencies, tolerances and the accuracy measures are common warehouse practice; the splits and examples here are starting points to adapt to your own data.