Measuring stock accuracy honestly
Most accuracy figures are calculated in the way that produces the highest number.
Ask a business how accurate its stock records are and the answer is usually a percentage with no definition behind it. The definition does most of the work, and there are several, producing very different figures from the same count.
Stock-accuracy records can be distorted by self-reporting bias when the person doing the work is also the only source of the measurement.
Three ways to measure the same thing
By line: what proportion of counted lines matched exactly? This is the strictest and the most useful for process, because it counts a variance of one unit the same as a variance of a hundred.
By unit: what proportion of total units were where they should be? Flattering, because a few thousand correct units of a cheap item drown out errors elsewhere.
By value: what proportion of stock value was accurate? The figure the accountant cares about, and it can conceal a badly inaccurate long tail.
Line accuracy tells you whether the process works. Value accuracy tells you whether the balance sheet is right. Quoting only one is how a business believes both are fine.
For a broader reference on product and location identification, GS1 standards document widely used supply-chain identification standards.
Decide the tolerance in advance
Exact matching is unrealistic for items measured by weight or counted in bulk. A stated tolerance — plus or minus a small percentage on bulk items, exact on serialised ones — makes the measurement meaningful.
Set it before counting. A tolerance decided after seeing the variances is a tolerance chosen to make the result acceptable.
Track the trend and the causes
A single accuracy figure is a snapshot. The useful version is a trend by month, and a breakdown of variances by cause — receiving, picking, unrecorded use, damage, unknown.
The unknown category is the one to watch. Where it grows, either the investigation is not happening or something systematic is being missed.
Set a target that reflects the cost of being wrong
Ninety-nine percent accuracy on every line is expensive and rarely justified. High accuracy matters where a stockout stops production, where the item is valuable, or where a miscount causes a customer failure.
For a long tail of cheap items, a lower target is a rational commercial decision rather than a failure — provided it is a decision.
Do not adjust silently
Every adjustment should carry a reason code and, above a threshold, an approval. Not because people cannot be trusted, but because the pattern of adjustments is the single richest source of information about where the process leaks.
A business that adjusts freely and records nothing has converted its most useful diagnostic into a monthly clean-up.