Most businesses with physical goods discover that the system's stock figure and the shelf's stock figure disagree. The causes are mundane and repeatable rather than mysterious.
Every transaction is a chance to diverge
Inventory records change when someone scans, keys or approves something. Each of those moments can record the wrong item, the wrong quantity or nothing at all.
Receiving is the largest single source. A pallet booked in as ordered rather than as delivered writes a fiction into the system that no later step will catch.
Because the error enters at the point of entry, downstream processes inherit it. Picking, selling and reordering all proceed confidently from a number that was wrong before anyone touched the goods.
Shrinkage is several different problems
Theft is the assumed cause and is real, but damage, spoilage, misplacement and administrative error usually account for a larger share of the gap in most operations.
These behave differently. Theft concentrates on small high-value items, damage concentrates on handling-intensive goods, and misplacement concentrates in warehouses where locations are not enforced.
Treating the whole variance as theft leads to security spending that does not close it. Separating the categories is the prerequisite for fixing any of them.
Units of measure quietly corrupt records
Suppliers ship in cases, businesses sell in units, and somewhere a conversion is stored. When a supplier changes case quantity without a system update, every subsequent receipt is wrong by a factor.
Similar problems arise with kits and assemblies. If the bill of materials does not match what the workshop actually consumes, component stock drifts steadily while finished goods look fine.
These errors are systematic rather than random, so they do not average out. They accumulate in one direction until a count exposes them.
Cycle counting finds errors sooner
An annual wall-to-wall count establishes accuracy once a year and tells nobody when the error started. Counting a small subset continuously spreads the work and shortens the detection window.
Items are usually prioritized by value and movement, so fast-moving expensive lines are counted often and slow, cheap ones rarely. The effort follows where errors cost most.
The purpose is diagnostic rather than corrective. A recurring variance in one location or one product family points at a process fault that adjusting the number does not repair.
Adjustments hide the underlying process
When a count disagrees with the system, the fastest resolution is to overwrite the record. That restores accuracy for a day and destroys the evidence of what went wrong.
Requiring a reason code on every adjustment turns those events into data. Patterns then become visible: one supplier, one shift, one product line generating most of the discrepancies.
The management question is not how large the variance is but whether it is understood. An unexplained gap of any size means the record cannot be trusted for purchasing decisions.