The Stock Count That Disagrees With the Ledger

Start by separating what each side is actually asserting, because the reason the reconciliation is so unsatisfying is that two different kinds of claim are being treated as the same kind of claim.

The count asserts a physical fact at a place and a moment. Four hundred and sixty-one units were on this bay, in this warehouse, under this light, at ten in the morning, counted by these two people who counted independently. That claim is scoped very tightly. It is true, and it is also nearly useless on its own, because it has no history. It cannot tell you which unit it was, when it left, who took it, or whether the same thing will happen again next month. The count is a photograph of one instant and the photograph has no caption.

The ledger asserts a running belief built from documents. It believes there were 480 units because it started from a figure and then added every receipt that was keyed and subtracted every issue that was keyed. Every one of those keystrokes was a claim about a document, not about a shelf. A goods-inward note says a truck arrived; it does not say the cases were put away, and if they were not put away the ledger is wrong in a way the count will detect but cannot explain. A stock-issue document says something left; it does not say it was carried out, and the difference between an issue recorded and an issue performed is where a great deal of stock goes missing.

So the count tests the belief and almost never replaces it. Where they agree, you have learned that the documents happened to match reality this time. Where they disagree, the count has told you the size of a gap and nothing at all about its shape, and the shape is the only part that is worth keeping.

Look at what a stock adjustment is. It is a movement document that reduces the ledger by the variance and carries a reason field, and the reason field is a free-text box that a tired person fills in at six in the evening. So the record that survives is: this much stock went somewhere, and the reason was shortage, which is not a reason. It is a restatement of the fact. The count sheet, which had names and times and bay numbers on it, gets photographed or not photographed and then is gone.

This matters more than it first appears, because of what the adjustment does to the next decision. Stock problems are not solved by adjusting. They are solved by finding which of maybe six ordinary things is happening, and the six are cheap to look for: goods that arrived and were not booked in, goods that left without a document, goods that were damaged and thrown away without a scrap note, a count taken badly, a movement booked to the wrong item, or stock sitting in the wrong place. Each of those has a different fix and a different owner, and each of them produces a variance of a similar size. You cannot tell them apart from the size of the gap. You can only tell them apart by looking, while somebody who was there still remembers.

There is a second cost, which is that the adjustment makes the ledger look like it was right all along. Once the nineteen units are written off, the stock figure is correct and the history is smooth and nobody can tell that the last three counts produced variances of 6, 19 and 23 units through three different mechanisms. A trend hidden inside a series of corrections is the most expensive thing a stock record can do, because it looks exactly like a well-run shop from the outside.

The 18% GST figure that appears in the worked example below is used only so that one purchase invoice total is checkable by hand. It is not a statement about the rate that applies to your business. How the tax component of a purchase invoice is treated in the value you carry for stock, and what your accountant needs to see alongside a variance adjustment, are questions for your own chartered accountant. Nothing in this article is tax advice, and we do not file GST returns or any other statutory return.

Here is a full count, constructed for this article, with every figure shown. One item, a 5 kg pack, sold in cases of twenty. Single warehouse, single bay, no manufacturing involved, no consignment stock, no stock sitting at a customer. These are invented figures chosen so the arithmetic can be checked by hand; they are not measurements of anything.

Opening position at 1 October, 08:00, per the ledger: 480 units. On 30 September two movements were keyed. Goods received against PO-318: 60 units, keyed at 11:20. Stock issued against SO-2201: 40 units, keyed at 16:05. So the ledger believed 480 units at 08:00 on 1 October.

The count began at 10:00 and was done twice, independently, by two people who did not see each other's figures. First pass recorded 457 units for the bay. Second pass recorded 461. The difference of 4 units was traced to a mis-add: one case of twenty had been recorded as two cases of nineteen, so 457 plus 4 is 461. One loose unit was then found in the aisle during a third sweep, and because it was found after the second pass it is counted separately below. Counted quantity for the count sheet: 461 units.

Variance against the ledger: 461 minus 480 is minus 19 units. The shelf is nineteen units short of what the ledger believes. At a landed cost of 640 per unit, constructed, that is 19 multiplied by 640, which is 12,160 of stock value that is not where the records say it is.

Then the part that matters. Before touching a single figure, somebody spends an hour with the people who were on the floor, and the nineteen units come apart into three named causes.

Same nineteen units, same twelve thousand one hundred and sixty, three completely different weeks of somebody's time. Nine units of that needs a counter rule and a rule on the repair job that closes only when a stock issue exists. Six units needs the scrap note made mandatory, the damaged units kept rather than binned so a supplier claim is possible, and somebody to look at whether the second bay is stacked the way it should be. Four units needs the counting method changed: count cases as cases, have both passes written down separately, and never overwrite the first pass with the second.

Write the adjustment off as one line and you have saved about an hour and destroyed the only chance to learn any of that. The adjustment is not wrong, and this is worth being clear about, because the argument is often heard as an argument against adjusting. Adjusting the ledger to match a verified count is the correct treatment. A count exists to make the ledger true, and a ledger that knowingly disagrees with the shelf is worse in every way that matters. The failure is not the adjustment. The failure is that the adjustment is allowed to be the last event, with no record attached to it of what the nineteen units were.

One more thing the constructed example shows. Two units turned up in the returns bin with a return note dated three days before the count, meaning the physical position was not 461 either, it was 463, and the count sheet was wrong about the shelf because stock was sitting in a bin that is not the shelf. If the two had simply been added to the count on the spot as found stock, the variance would have been recorded as minus 17 instead of minus 19, and the count sheet would still be wrong. This is why a count needs a named sweep order and a defined scope, and why a bay that is not part of the count cannot quietly contain counted stock.

The test of a count is not whether the number matched. It is whether the record you kept afterwards would make the next count cheaper to investigate. A count that produces a number and nothing else has cost you a full day of counting and one afternoon of adjusting, and it returns the same blank answer next quarter. A count that produces a number plus three named causes plus one change to the process returns an answer that is narrower each time, and the goal is not zero variance, which is not an achievable thing for a business that handles physical goods, but a variance made of causes you recognise.

Six things have to be captured, and none of them is the adjustment. The count needs a defined scope written down before anybody starts: which bays, which item, which locations in and out, and what time window is excluded because a movement is legitimately in flight. It needs both passes recorded separately, with the reconciliation between them shown, because a single figure conceals exactly the counting-error cause that this article's example found for free. It needs the variance decomposed into named causes, not one total, and a cause that is honestly unknown must be allowed to be recorded as unknown rather than being given a plausible name. It needs the person who investigated and the date, so that next quarter somebody can ask them what they saw instead of guessing. It needs the adjustment to point at the count rather than replacing it, so the count sheet stays readable. And it needs one line about what will change as a result, with an owner, because a variance record that changes nothing is a record nobody will keep writing.

The obvious next move after a variance like the one above is to divide the short units by the units you believed you had, and to publish the result as your shrinkage rate. We are not going to print that number, and the reason is not squeamishness. A printed shrinkage figure is a number that readers will compare against a shrinkage benchmark, and we have never measured a benchmark, so any comparison you make with it would be noise dressed up as a finding. More importantly, the ratio would not change anything. Nineteen units at four percent and nineteen units at half a percent are the same nineteen units with the same three causes, and the causes are what you can act on. We would rather give you the decomposition and no percentage than a percentage and no decomposition.

There are two further things a count structurally cannot deliver, and both of them are commonly promised. It cannot tell you about stock you did not count: goods at a customer, goods in transit, goods in a returns bin, goods in a godown you visit once a year. A count is a statement about scope and the scope is whatever you wrote down. And it cannot tell you the value of a variance without an agreed cost, which is a decision, not a measurement, and a decision that belongs to your accountant rather than to a counting form.

What a count can do, and what we think is worth doing properly, is narrow and durable: turn an unexplained number into a small number of named causes, each with an owner, so that the same cause stops recurring quietly. That is a lower claim than an accuracy percentage and it is the only one that survives contact with a busy week.

Stock movement, purchasing, warehouses and low-stock signals live in the Commerce area of the platform. That means a receipt is a document that moves stock, an issue is a document that moves stock, a warehouse is a place the stock is held in, and a low-stock signal is a number crossing a threshold you set. Every one of those is a record rather than a number typed into a spreadsheet that is reconciled by hand at month end. Expenses are not part of that: expense claims and staff costs live in Nox-Billings only, and Nox-Billings does not own your stock. An invoice and a payment are two different records, and paid is a projection of allocations rather than a column you set, so a purchase invoice marked settled is a derived state you can always walk back to the receipts that produced it.

Here is the list of things this system does not have, stated here rather than left for you to assume. There is no demand-forecasting model, so nothing here predicts what you will sell next month. There is no automatic purchase-order engine, so nothing here will raise a purchase order on its own because a number crossed a threshold; a low-stock signal is a signal, and the order is a decision somebody makes. And there is no manufacturing capability set: no bill of materials, no shop-floor scheduling, no production costing. If your stock is consumed by making things rather than by selling them, that is a different category of software and we would rather say so than imply that a warehouse record covers it. There is no change-request record, no milestone record and no deliverable record, no contract editor and no e-signature, so a scope change is a new quote raised against the same project. There is no timesheet record and no payroll module, and we do not file GST returns or pursue statutory recovery.

Two more, because stock pages are exactly where people expect them. Duplicate detection flags candidates and a person reviews them: match rules and merge behaviour are a setup decision and nothing merges automatically. And shift close and day-end reconciliation are assisted-setup maturity rather than a switch you flip yourself, because the counted cash, the expected cash and the variance only reconcile properly once somebody has thought about your till float and your shift boundaries.