The sampling method in a revision

A wholesale furniture-fittings warehouse holds 6,000 positions: handles, hinges, drawer runners, fasteners. Counting everything would take five people six days, and the manager has no such time. This is where a sampling revision helps: part of the positions are counted and the result lets you judge the state of the whole warehouse. But "we'll count whichever shelves are handy" is not a sample, just reassurance. A sample must follow clear rules.

What a sample gives and what it does not

It assesses how reliable the records are: whether the share of errors is small or large. It points to problem groups and saves time. But it does not give the exact amount of the whole shortage. Nor can a shortage on an uncounted position be written off officially, since it was never actually checked. Whether a sample result can stand in for an official stock count, find out in advance from an accountant or lawyer.

How to build a sample

1. Sort the positions by value, from the most expensive to the cheapest.
2. Count the expensive layer in full. A few hundred positions usually make up most of the total value.
3. Choose the rest at random. In Excel, add a column of random numbers beside the list, sort, and take the number you need from the top, or mark every thirtieth row. Not the convenient shelves.
4. Check that different zones and slow-moving goods have landed in the sample.
5. Check both ways: from the list to the shelf (find the chosen position on the rack) and from the shelf to the list (take twenty random items off the shelves and see whether they are on the list). The second finds goods that were never booked.

The larger the sample, the more reliable the result. There is no rigid formula; as a practical guide, take two or three hundred positions from the random layer, or roughly a quarter of the total in a small warehouse. Do not draw conclusions from twenty or thirty positions: a single chance error would visibly swing the result.

An example: 500 positions out of 6,000

The expensive layer: 300 positions, about 70 percent of the value. They were counted in full and differences appeared in 11 (3.7 percent). From the remaining 5,700, 200 were chosen at random: differences in 14 (7 percent). The threshold had been set in advance at 3 percent (an illustrative figure), so the records were judged unreliable.

But split by group, 9 of the 14 differences turned out to belong to drawer runners: they are sold in pairs, yet were booked singly. So not the whole warehouse but only the runner group was recounted in full, and the unit in the records was corrected. The sample took two days and the group count one, instead of six days for a full count.

Mistakes in sampling

• choosing from a "convenient" place: the shelf by the door or a zone recently tidied;
• a sample that is too small: twenty positions prove nothing;
• errors cluster in one group and a random sample misses it: look at the result by group as well;
• "multiplying" the result up to the whole warehouse and writing it as an exact shortage;
• stopping when the error share is high: it is not a verdict but a signal to count more deeply.

When a sample suits

A sample works well as an interim check: between quarters, after a new storekeeper starts, or before a big revision, to learn how clean the records are. Do the annual closing count in full.

Call a sample a "sample" in the document, not a full revision, so that no false confidence in the result arises later. Keep the list of chosen positions and how they were chosen: if someone asks "why these?", you will have the answer. A sample should be repeatable under the same rules, so that results can be compared with each other. If each quarter's sample result goes into the revision log, you can see in advance which way the error share is moving.

Conclusion

A sample saves time, but only with clear rules: the expensive layer in full, the rest at random, a threshold in advance, the result by group. The Reviziya team can help you plan a sample and read its result correctly.