Counting the whole plant without stopping it

The annual stock take is one of those practices that survives because nobody asks what it is for. It stops production, occupies everybody who knows anything, produces a variance nobody can explain three weeks later, and delivers an accurate stock figure for exactly one day — after which it decays at the same rate it did before.

The alternative is not counting less carefully. It is counting more often and smaller, so that no single count needs the plant to stand still, and so that a variance is found while somebody still remembers the week it happened in.

Count by class, not by calendar

Rank items by annual consumption value and the usual thing is true: a small minority of codes carry most of the value. Count those often — monthly, or even weekly for the handful at the top. Count the middle quarterly. Count the long tail once a year, and accept that you are counting it mostly for completeness.

The point is not the letters A, B and C. It is that a fastener bin and a batch of platinum catalyst do not deserve the same attention, and a calendar-driven full count gives them exactly the same attention.

Add a second trigger alongside value: count anything that just behaved oddly. A stock-out on an item the system said was available. A negative balance. An item whose consumption jumped without a matching order. Those are the counts that find something.

Cut-off is the whole game

A count is a statement about a moment. Everything that happens after that moment belongs to the next side of the line, and the single most common way a count goes wrong is that this is never decided precisely.

Practically, that means:

Freeze the bin, not the plant. While a location is being counted, no issue and no receipt against it. Everything else in the plant carries on.
Material physically present but not yet received — sitting at the gate, awaiting inspection — is either in or out, and whichever you choose, it must be the same choice in the count sheet and in the system.
Material issued to the floor but not yet consumed is not in the store. It is WIP. If your count sheet and your WIP register both include it, you have counted it twice and your variance is a fiction.
Nothing gets posted against the counted location until the count is closed.

WIP is where counts go to die

Store stock is easy — it is in a bin, in a package, with a label. Work in progress is material in a machine, in a queue, in a bin at the end of a line, in a bucket of parts waiting for the next operation, and at a job worker’s premises three districts away.

The only reliable way to count it is not to count it as material at all. Count it as work against a job card at a stage: 340 pieces at operation 4, which by the routing represents so much of each input. That converts a counting problem into a reporting problem, and reporting is something the floor can do daily rather than annually.

Which is really an argument for stage-wise progress capture existing before the count does. If you do not know what is at which operation on an ordinary Tuesday, no count will tell you on a Sunday.

The variance is the deliverable

Adjusting the system to match the count is the least useful thing you can do with a count. The number is not the output; the reason is. Classify every variance past a tolerance:

Issued and never booked — a paperwork lag, usually the largest bucket and the cheapest to fix.
Booked in the wrong unit of measure — kg against pieces, metres against rolls. Look for variances that are suspiciously close to a conversion factor.
Received without a GRN, or against the wrong code.
Scrap and process loss generated but never recorded.
Wrong item picked — which shows up as two variances, equal and opposite, in neighbouring bins.
Genuinely missing.

Set a tolerance per class — tighter for high-value, looser for fasteners — and route anything past it for approval rather than letting a counter adjust it. An adjustment without a reason code is not an adjustment; it is a cover-up with a timestamp.

What good looks like

Counts happen weekly, take an hour, and nobody stops. Each one names its locations and its cut-off moment. Variances carry reason codes and the recurring reasons get fixed at source, so the variance rate falls year on year — which is the actual measure of whether any of this is working. And the annual count, if your auditor still wants one, becomes a verification of a system that already works rather than the only time anybody looks.

What MANUPRIME does about it

Physical count and cycle count are both in the inventory module, against multi-store, multi-bin locations with batch and expiry, so a count can be scoped to a bin without freezing anything else. Variances are posted with a reason and land in the record’s activity trail, and WIP is tracked stage-wise against the job card rather than as a lump of material, so the floor is not counted twice.

Inventory and WIP tracking are both screens we walk through — ask for a demo.