A Count That Explains Itself
Most stock counts produce a new number and destroy the evidence of how the old one went wrong. That is a smaller thing than it sounds, and it costs you the same amount every year.
A count happens. Somebody walks the shelves, the figures come back, and where they differ from the system the system is corrected. The count is now accurate. Everybody moves on.
And the single most useful piece of information the exercise generated has just been thrown away — not the new number, which you would have got to eventually anyway, but the answer to how the old one went wrong. That answer is the only part that changes anything next quarter.
A count that produces an adjustment tells you the size of a problem. A count that produces an explanation tells you where it comes from. The difference between those two outputs is almost entirely a matter of what was recorded while counting, and it is the cheapest improvement available to most operations.
This is a process argument, not an accusation
The overwhelming majority of variance is process rather than dishonesty: a transfer nobody confirmed, a unit of measure applied differently by two people, goods received against the wrong line, a return that went back on the shelf without a document. Any post about counting that implies theft is a post nobody will show their warehouse manager, and the manager is exactly who needs to read it. There are no shrinkage benchmarks here either — there is no honest industry figure to quote, and an invented one would undermine everything else.
What an explanation actually requires
Four things, none of them difficult, all of them decided before the count rather than after it.
-
It is blind
The counter does not see the expected figure. Once they do, the count measures agreement with the system rather than what is on the shelf — and it will agree, because human beings are obliging. This single property determines whether the exercise is evidence or theatre.
-
The variance is valued
A difference of four units means nothing on its own. Four units of one thing and four of another are not the same finding, and a count that produces quantities rather than values cannot be prioritised.
-
It belongs to a place
A variance attributed to the whole business is an interesting number. A variance attributed to one location, bin or van is an instruction — and the pattern across places is usually the finding.
-
It belongs to a period
Which is what turns a count into a series. One count is a fact; four counts are a trend, and a trend at one location is the thing you can actually act on.
Blind, valued, placed and dated. Miss any one of the four and you get a corrected figure instead of a reason — and next year the same correction, of about the same size.
The upstream causes a good count keeps pointing at
When variance is attributable, the same handful of causes turn up. That is the point: they are fixable, and they are only visible because the count recorded where and when.
| What the pattern looks like | What is usually behind it |
|---|---|
| One location, consistently short | Movements out that were never recorded, or recorded to the wrong place |
| Two locations, equal and opposite | A transfer despatched and never confirmed on arrival |
| One product, everywhere | A unit-of-measure or coding problem rather than a stock one |
| Worse after a busy period | Receiving under pressure — goods accepted before they were matched |
| Only at a remote site | Capture that requires a connection the site does not reliably have |
What is ours, and where it stops
Blind counts with valued variance
The counter does not see the expectation, and the output is a valued difference attributed to a location and a period rather than a quantity to be adjusted away.
Transfers confirmed on arrival
Which removes one of the largest causes of variance before any count happens: goods that left one place and were never acknowledged at the other.
Coded items with a real unit of measure
Classification and unit as properties of the product rather than choices made by whoever is keying — the fix for the one-product-everywhere pattern above.
Offline capture at remote sites
Stock issues, returns, transfers and asset movements recorded without a connection and synced on return. Note the limit honestly: counting itself needs a connection, so a count at a site with no coverage is a connected activity and worth planning for.
Telling you why the variance happened
Not ours. We make it attributable — to a place, a period and a value. The pattern is visible; the cause is in your operation and your people know it far better than any software does.
The question to ask about your last count
Not what the adjustment was. What it was attributed to.
- Was the counter able to see the expected figure?
- Is the variance available by location, or only in total?
- Is it valued, or only in units?
- Can you compare it to the previous count at the same location?
- Did anything change in the operation as a result — and if not, what was the count for?
That last question is the honest one. A count that changes a number and nothing else is a compliance activity with a cost and no return. It is also, in most operations, the only kind of count anybody has ever done — which means the improvement available here is not marginal.
What is built here, what is not, and what we would decline is on the Namibia market page — including the one item in its honesty ledger that limits the page's own argument. The money side of this market is your import VAT is not calculated on the invoice.
