Cycle counting is the practice of counting a small, rotating slice of your inventory on a schedule, a few SKUs today, a few tomorrow, so that over a cycle everything gets verified without ever freezing operations for a full count. For online sellers the point is sharper than accounting hygiene: your channels advertise your recorded counts, and records that drift from reality oversell with perfect confidence.
Why cycle beats annual
The annual wall-to-wall count fails on three fronts: it freezes selling (expensive), finds errors up to a year old (unfixable, cause long gone), and treats every SKU as equally worth counting (false). Cycle counting inverts all three: no freeze, errors found weeks old at most, and attention allocated where errors cost most.
Building the rotation
Weight by class. Using your ABC ranking: A-items counted monthly or better, B quarterly, C once or twice a year. The A-item counts protect the revenue; the C counts exist mostly to keep the tail honest.
Weight by error-proneness. Some inventory generates discrepancies structurally: multi-location SKUs, look-alike variants, items sold loose from cases, high-touch picking zones. Promote chronic offenders to a faster lap regardless of class.
Make it boring. Ten minutes daily or one hour weekly, same time, a count sheet generated from the system (blind counts, the counter should not see the expected number), and one person accountable per week. Rituals survive; projects die.
The discrepancy playbook
A count that disagrees with the record is information, and the response order matters:
- Recount before adjusting, a second look resolves a surprising share.
- Check the timing windows: an order picked but not yet shipped, a return received but not yet restocked, a sync event in flight. The record may be right and mid-motion.
- Adjust the record and log the reason, adjustment without a reason code is how the same error returns monthly.
- Chase the pattern, not the unit. Three discrepancies in the same picking zone or SKU family is a process defect, mislabeled bins, variant mix-ups at receiving, a bundle SKU decremented wrong.
Track one metric: inventory record accuracy (counts matching the record within tolerance, as a percentage). Trending it per class tells you if the process is working; the industry-respectable bar is in the high nineties for A-items.
Multichannel notes
- Adjust in the source of truth, once. With a hub keeping channels synced, a corrected count propagates everywhere in seconds; hand-editing per channel is how corrections create new drift.
- The record you count against must be the live one. Counting against a stale export produces phantom discrepancies; count against the system’s current number at the moment of counting.
- Discrepancy direction is a clue about the channel layer too: recorded-higher-than-physical repeatedly on synced SKUs can indicate an order flow (a channel or manual sales lane) not wired into the pool.
Common questions
How many SKUs should I count per session?
Small enough to be sustainable, ten to thirty for a small operation. Consistency beats batch size by a wide margin.
Do I still need an annual full count?
Some accountants and jurisdictions want one; many accept a demonstrated cycle program with accuracy records. Ask yours, and keep the discrepancy log either way, it is the evidence.
Blind counts or verified counts?
Blind (counter sees no expected quantity). Knowing the target number converts counting into confirming.
What tolerance should trigger investigation?
Zero for A-items and serialized goods; small percentage tolerances only where unit costs are trivial and miscount cost is noise.
Count a little, trust a lot
Live counts across every channel, adjusted once and synced in seconds, from $49/month with unlimited orders, with the per-SKU velocity to weight your rotation on the plans above. See pricing.