Every retailer has had the conversation. The system says forty-two. The shelf has thirty-seven. Nobody can explain the five, so the system is adjusted to thirty-seven and everyone moves on.
Repeated across a few thousand lines, that adjustment is the single largest unexamined cost in most retail operations — and the reason it stays unexamined is that the discrepancy is treated as one problem with one cause, usually assumed to be theft.
It is at least six problems. They occur in different places, at different rates, and only one of them is theft. Counting more often finds the gap sooner; it does not close it.
Where the units actually go
Receiving that was never verified. A delivery of ninety-six recorded as a hundred because the docket said a hundred. This is the largest single source in most operations and the least suspected, because the error enters the system as an authoritative document.
Sales recorded against the wrong item. Two variants at the same price, and the counter scans whichever barcode is nearest. Revenue is correct, stock is wrong for both lines, and nothing in the day's takings reveals it.
Returns handled inconsistently. A return that goes back to the shelf without being recorded, or is recorded and left in the back room. Both create a discrepancy in opposite directions.
Damage and expiry removed informally. A broken item swept up, a expired unit discarded. Entirely legitimate, rarely recorded, and cumulatively substantial in categories with short shelf life.
Internal consumption. Staff use, samples, display units, items given to a customer to resolve a complaint.
Theft. Real, both external and internal, and typically a smaller proportion than the first five combined.
The reason the distinction matters: five of the six are process failures with cheap fixes, and one is a security problem with an expensive one. A retailer who assumes theft installs cameras and continues to lose stock at receiving.
Why the annual count cannot fix it
A full count once a year produces a single number: how wrong you were over twelve months. It cannot tell you where, when or why, because every cause has been averaged together over a year of trading.
It also arrives too late to act on. Whatever process failure caused the loss has been running for months and will run for months more.
The alternative is a rolling count — a small number of lines counted every week, chosen to cover high-value and high-movement items more often than slow ones. The total counting effort is comparable. The output is completely different: a discrepancy found within a week is attributable to a specific period, a specific delivery, and a specific set of shifts.
That attribution is the whole value. An unattributed variance is a write-off; an attributed one is a fixable process.
Fixing receiving first
If only one thing changes, it should be this.
Count what arrived, against the order, not against the docket. The docket states what the supplier intended to send. Those are different claims, and accepting the second as evidence of the first is how the largest share of variance enters.
Record short deliveries at the moment they are found, while the claim is still credible with the supplier. A shortage discovered three weeks later is absorbed rather than claimed.
Give receiving to someone with time to do it, which frequently means not during peak trading hours.
Every unit that enters the record incorrectly at this point is wrong for the rest of its life in your system, and no amount of downstream counting recovers it. The systems view of this is in what actually prevents stock variance, and for distributors specifically in warehouse management for growing distributors.
Making the right entry the easy one
The second structural fix is at the counter, and it is a design problem rather than a discipline problem.
If recording a damaged item takes six taps and a manager's approval, it will not be recorded. If two variants sit adjacent with near-identical packaging, the wrong one will be scanned during a rush. If a return requires a supervisor who is not present, the item goes back on the shelf unrecorded.
Each of those is fixable by making the correct action the fastest one available — which is the general principle behind what breaks at peak hour. Staff overwhelmingly do the right thing when the right thing is not slower than the alternative.
The number worth watching
Total variance in currency terms is the figure most retailers track and the least useful, because it moves with turnover and hides everything.
More informative:
Variance by product category, which localises the cause. Short-shelf-life categories behave differently from electronics.
Variance by outlet, which is the strongest available diagnostic in a chain — one outlet consistently worse than comparable others is a specific, findable process difference rather than a general problem. That comparison depends on common definitions, which is the argument in keeping every outlet on one truth.
Variance by direction. Consistent shortages and consistent surpluses have different causes. Surpluses usually indicate recording errors rather than anything else, and they are informative precisely because nobody steals stock onto your shelves.
Adjustment frequency, which reveals how often the system is being corrected rather than how large the corrections are.
Frequently asked questions
What variance level is acceptable?
Published benchmarks vary enough by category and format to mislead. Your own trend is the meaningful measure — establish a baseline, then watch direction and attribution rather than a target figure.
Should we count everything or sample?
Sample, weighted toward value and movement. A weekly rolling count of the lines that matter beats an annual count of everything, because it produces attributable results.
Is it theft?
Usually less than assumed. Establish receiving accuracy and recording discipline first — if variance persists after both are sound, the remaining figure is a much better basis for that conclusion.
Does barcode scanning solve it?
It removes one category — misidentification at the counter — and does nothing for receiving, returns, damage or internal use. Necessary, not sufficient.
Who should count?
Not the person responsible for the section, where that is avoidable. Not as an unpaid extension of a full shift, where accuracy collapses. Counting done badly produces data worse than no count.
Where to start
Take your ten highest-value lines and count them this week. Then check the last delivery of each against the purchase order rather than the docket.
Most retailers find at least one discrepancy at receiving, which means it entered the system wrong before anyone had a chance to lose it — and that is the cheapest variance you will ever recover.
Where stock accuracy sits among the other retail technology decisions is in choosing technology for your sector.
If you want stock accuracy treated as a process problem rather than a counting problem, get in touch.