Full Count or Sample Count? Choosing the Method for the Decision

A full count of everything in a warehouse is a big job that takes days, has its cost and affects the running of the operation. A sample count is faster and cheaper, but it answers narrower questions. So the question that should come before the choice is: what decision is waiting for the result? If the decision is a deal worth millions, a sample may not be enough. If it is a matter of knowing whether things are in order at a partner's warehouse, it may be more than enough. What follows is how to match the method to the decision, and how to explain honestly what it covered.

When a full count is needed

A full count suits situations in which a large commitment rests on the figure or a dispute is to be settled by it:

• Buying or selling a business, or buying a stake in one.
• Handing over and taking over a warehouse when the operator changes or the contract ends.
• A dispute between two parties over quantities.
• A suspicion of a large loss.
• Goods that stand as security for a loan.
• Small stock: counting all two hundred items costs not much more than counting a sample of them.

When a sample is enough

A sample suits periodic oversight, finding out a trend, and large stock with many items of low value. It is also right when you want to decide whether a full count is worth doing. If the sample comes out clean, you may be reassured; if problems show in it, that is a reason to widen. It does not replace a full count in acquisition decisions, but it shows early whether one is called for. A sample may be taken quarterly, with the selection changed each time, so staff cannot know in advance what will be checked.

How a sample is built

A good sample is not chosen by the owner of the warehouse. There is a practical way to build it:

1. Rank the items by value, from the largest to the smallest.
2. Count the top band in full: the items of greatest value, because one error among them equals many errors among the small ones.
3. Choose from the rest at random: for example every line after a number drawn by lot, and leave the choice to nobody.
4. Add what calls for caution: goods with expiry dates, items that are easy to take, and any location that raised a doubt.
5. Write down the sample size and the selection rule before you start.

The sample size in statistical terms is set by a specialist if needed; the practical rule is that coverage rises as the decision grows.

A worked example (conditional figures)

A warehouse has 2,000 items worth 1,500,000 dollars. The largest 100 are worth 975,000 dollars, 65 percent of the value. The plan: count all 100, and count 100 random items from the other 1,900. The estimated effort is three days for a full count and one day for the sample. The result: 3 of the 100 large items differ, and 7 of the 100 random ones. How do we read this? The coverage by value of 65 percent is complete, with a sample of the rest. We do not say that "7 percent of the stock is wrong", which is a conclusion larger than the evidence. We say that in a sample of 100 small items 7 differed, that this suggests control over small items is weaker, and that if the decision depends on them, they deserve a wider count.

Explain the coverage honestly

A phrase such as "stock accuracy is 98 percent" sounds convincing, but it claims what a sample cannot allow. Write instead:

• What was counted in full, what by sample and what not at all.
• How the sample was chosen, and by whom.
• The coverage by value and by number of items.
• What can be concluded and what cannot.

A reader who sees these details trusts the report more than one given a single shiny number. And if an investor wants a figure for all of the stock, he or she needs the full count, and a statistical approximation will not do.

In short

Choose the method by the decision: a full count for deals, handovers and disputes, a sample for periodic follow-up and finding the trend. WarehouseDepo helps you match method to decision, writes in its report what the sample covered and what it did not, and widens the count where the differences point to it.