Almost every warehouse claims 98% inventory accuracy. Almost no warehouse at 98% actually runs well. The contradiction is not a lie: that 98% was calculated on total units, the most generous of the three ways of measuring. Measuring by unit, by item or by location produces three different numbers for the same warehouse, and only the last one resembles what happens when somebody goes to pick goods and they are not where the system says they are.
The three ways of measuring accuracy
Accuracy by unit: you add up the units counted, compare them with the system figure and take the percentage. It is the most common metric and the most forgiving, because a shortage of three units against an item holding twenty thousand dissolves until it disappears. Accuracy by item: you count how many items closed exactly against the total number of items counted. Each item is worth one, whether it holds twenty thousand units or five. It is considerably more demanding and it is the one that serves purchasing and planning. Accuracy by location: you count how many locations held exactly what the system declared, meaning the right item, the right quantity and nothing else, against the total number of locations audited. It is the harshest and the only one that represents the operator experience, because an operator does not look for the item stock: he looks for that box in that position.
Why accuracy by location is the one that hurts
Value recovered time using your actual operating costs and measurements of the tasks that change. Distinguish time available for other work, reduced overtime and expenses that can actually be avoided.
How is the inventory accuracy rate calculated?
The formula is always the same: correct measurement units divided by total measurement units audited, times one hundred. What changes everything is what you choose as the measurement unit. If the unit is the physical unit, you add absolute discrepancies, not net ones, and subtract them from the total. That nuance matters: a shortage of five in one location and an overage of five in another do not cancel out. They are two errors, not zero. If the unit is the item, each item counted is worth one and only adds up if it closed exactly. If the unit is the location, each location audited is worth one and only adds up if it contained exactly what was declared. Three rules make the number useful: it is calculated on what was audited in the period and not on the whole catalogue; it is calculated before the adjustment and not after; and it is always published with the same measurement unit, so one month can be compared with another.
What counts as a discrepancy and what does not
A discrepancy is any gap between what was declared and what was found. There are four cases, and the fourth is the one usually forgiven for no good reason. Shortage: there is less than the system says. Overage: there is more. Wrong item: the quantity is right but the article is not. And wrong location: the goods exist and are correctly counted, but they are not where the system sent someone to find them. Many companies do not record the fourth case because the stock is there. It is precisely the error that breaks order picking: if your metric does not count it, your metric is not measuring the problem you have. Adjustments made during the count do not qualify as correct either. Accuracy is calculated with the figure prior to adjustment; otherwise every warehouse measures 100% every month and the series tells you nothing.
Why 98% coexists with an operation that fails every day
Take a warehouse with two thousand active locations and accuracy by unit of 98%. That deviation, spread out, can mean dozens of locations holding something wrong. Each one generates an interruption when somebody goes to pick it, and interruptions are not evenly distributed: they land on the items touched most often, which are exactly the ones that move every day. That is why 98% by unit and a sense of chaos coexist without contradiction. The metric is averaging over a large, static stock; the operation is tripping over the small, active part. The way out is not to change the number, it is to change the measurement unit. When you start publishing accuracy by location, the percentage drops sharply and is embarrassing for a few months. That lower number is the one you can improve, because it is the only one that corresponds to something happening on the floor.
What accuracy an automated system reaches
Inventory accuracy is measured by comparing records with physical stock. Results depend on receiving and dispatch records, location rules, training and controls; software or robots do not guarantee an accuracy percentage.
When chasing the decimal point is not worth it
Priorities depend on the consequences of an error. A small discrepancy in a critical part may stop an operation, while a larger difference in another item may be resolved through routine replenishment. Segment by value, criticality and discrepancy frequency. Set explicit tolerances rather than hiding them inside an average, and focus counts and recording improvements where operational risk is greatest.
What to measure from Monday on
Choose one measurement unit and stick with it. If you can, make it the location: it will hurt at first and it is the only one that shows you where the operation breaks. Calculate on what was audited, before the adjustment, and publish the number with its definition alongside so nobody compares it against something else. With three months of that series you will know whether you have a process problem or a physical limit. Only then is it worth discussing equipment, and you will be able to do it with your own data.
Frequently asked questions
It depends on the measurement unit, so the question cannot be answered with a loose number. 98% by unit is common and says very little; the same warehouse measured by location usually reads considerably lower. Before setting a target, define the unit, calculate before the adjustment and hold that definition for the whole year.
No. Discrepancies are added in absolute value, never netted. A shortage of five units in one location and an overage of five in another are two distinct errors with two distinct causes, and offsetting them removes from the indicator exactly the two problems you were supposed to investigate.
Accuracy by unit compares total quantities and dilutes small errors within large stocks. Accuracy by location asks whether each audited position held exactly what was declared. The second is harsher and represents what the operator experiences, because he looks for a box in a position, not for a system total.
Inventory accuracy is measured by comparing records with physical stock. Results depend on receiving and dispatch records, location rules, training and controls; software or robots do not guarantee an accuracy percentage.
With daily cycle counting, the indicator is recalculated monthly on what was actually audited in that period. Publishing it more often makes it noisy, because a small sample varies a lot; publishing it once a year makes it useless, because by the time the number arrives there is no room left to correct anything.