Count sessions: taking stock without stopping the warehouse

Saturday morning, 7:30. The warehouse lead is standing at the head of aisle two with a printed sheet of 312 lines. The courier pickup is at nine; the marketplace cut-off for same-day picking is at eleven. The instruction from management is unambiguous: nothing moves until the count is finished. By noon barely half the sheet is ticked, two people have counted the same rack twice, the picking team has been idle at the door, and three orders have slipped to Monday. The thirty-line variance list that appears at six in the evening tells nobody which variance is a genuine shortage, which is a miscount, and which is simply a transfer that moved through the building that day.

This article is about never having that Saturday again, by treating a count as a session rather than a day. We start with why stopping the warehouse is the most expensive option available, then look at what a count session technically freezes and what it deliberately does not. From there we cover how to split the count sheet, why the person counting should not be the person approving, and the order in which to investigate a variance. The closing sections deal with how variances land in the books, the single number that measures counting accuracy, and how to build the routine in your first ninety days.

Why is counting without shutting down possible at all?

Full physical counts are a habit inherited from paper: a printed sheet, a frozen moment, and numbers keyed in once at the end of the day. In that world stopping the operation is mandatory, because once the sheet is printed the system has no record of what happened in the building. The moment the record becomes digital, that constraint disappears. If the system knows the quantity at the instant the sheet opened, and logs every receipt, issue and transfer after that instant separately, it can split the gap between counted and on-hand into two parts: a real variance, and legitimate movement that occurred during the session.

The practical consequence is that you can count while shipping continues. The real gain is not one rescued Saturday — it is turning counting from an annual ordeal into a weekly habit. A business that counts once a year makes promises on wrong numbers for eleven months. A business that counts weekly catches an error within a few days on average. Getting inventory management to the point where you neither tie up cash nor run out depends on data being corrected continuously; one large annual count cannot do that.

What does a count session actually freeze?

What a session freezes is not stock, it is the reference. For every line in scope, the session stores what the system believed the on-hand quantity was at the moment the sheet was generated. Physical goods keep moving, and the available quantity in the system keeps changing. The only thing held still is the baseline that the counted number will be compared against.

Rocketly's Count Session Detail screen makes that split visible. You generate the sheet by warehouse and ABC class, enter counted quantities line by line, and see the variances on the session screen as you go. If a legitimate stock movement hits one of those lines after the session opened, the adjustment that will be applied on completion is shown separately. In other words the screen gives you two numbers: the gap between what you counted and the frozen baseline, and the sum of movements that happened after the sheet opened. Mixed together they destroy confidence in the count; kept apart they turn the count into evidence.

Does movement after the sheet opens ruin the count?

No — provided that movement is recorded. That is the single real precondition for counting without downtime: everything that happens in the building during the session must reach the system. If a picker empties a shelf and plans to raise the delivery note that evening, the quantity is simultaneously gone from the rack and present in the system, and it will surface as a variance. So the question to answer before you switch to continuous counting is not "which day do we count" but "how many minutes late are our movements recorded".

The cheapest way to push that lag toward zero is scanning. Running a warehouse on barcodes for faster, error-free counts does more than speed up counting; it moves the moment of recording closer to the moment of the event. In Rocketly, giving every product an SKU and a barcode reduces both Barcode & SKU lookup and finding the right line on a count sheet to a single scan. A counter searching by eye writes into the wrong line; a counter scanning cannot.

How often should each item be counted?

Counting everything at the same frequency wastes effort and hides real risk. Frequency should follow two properties: the value of the item and how fast it moves. On a high-value, fast-moving line a single-unit error carries both cash and promise risk; on a low-value item that moves twice a year the same error is close to irrelevant. The right interval for a cycle count sits at the intersection of those two axes.

The thresholds below are a starting frame, not a rule — you are expected to tighten or relax them once you can see your own variance rates. For seasonal lines, demand rather than the calendar sets the cadence: pre-season and in-season frequencies should differ. Teams that align demand forecasting with stock and sales data in one place know which items matter before the peak arrives, not after.

Item class	Typical profile	Suggested interval	Acceptable variance
A — high value, fast moving	Large revenue share, weekly movement	Every 4 weeks	Under 0.5% by units
B — mid value or mid velocity	Regular monthly movement	Every 12 weeks	Under 1% by units
C — low value, slow moving	A few movements per year	Once or twice a year	Under 2% by units
Lot / serial tracked	Items with expiry dates	Every 4 weeks, by batch	Zero tolerance
Newly created SKU	First 90 days	Twice in month one	Zero tolerance

Should the sheet be split by aisle or by item class?

Both work, and they solve different problems. Splitting by aisle shortens the walking distance and speeds the count up, but you need several sheets before you can say anything about a whole class. Splitting by class makes accuracy manageable: when a session closes you can state that every A item has been counted. In practice the most efficient approach combines them — pick the class, then walk it in aisle order.

That means classification has to exist before the sheet does. Prioritising stock by value through ABC analysis is an input to the counting plan, not an output of it. Because Rocketly's ABC Analysis already splits stock into value classes, the count sheet can be generated straight from that class, and the next sheet reflects the new distribution automatically when the classes shift.

The second split criterion is dormancy. An item nobody has touched for months is less urgent than a fast mover but produces bigger surprises, because the error has been accumulating for years. Turnover and aging analysis flags those lines, and you guarantee that every one of them reaches a sheet at least once a year.

Why separate the counter from the approver?

Counting is by nature the process most exposed to covering its own errors. If the same person counts and approves the variance, looking at the system quantity and writing it down — what the trade calls a non-blind count — becomes inevitable. The simplest control is not showing the system quantity on the counter's sheet. The second is giving approval of the adjustment to a different role.

For that separation to mean anything, records must not be editable after the fact. In Rocketly, receipts, issues, transfers and reservations sit in immutable stock movement records: when a correction is needed the original line is never deleted, an adjusting movement is added against it. Count sessions follow the same logic — the variance is posted as a separate adjustment at the moment of completion. Three months later, the question "where did those 40 units go" still has an answer, with a date and a user attached.

Investigate first or adjust first?

Recount first. A meaningful share of variances are counting errors: writing into the wrong line, not knowing how many units are in a carton, missing that the same item also sits in a second location. So the working rule is that no variance above the threshold becomes an adjustment until a different person has counted it again. If the second count confirms the first, the variance is real and the cause is worth chasing.

An efficient order for chasing it: movements during the session first, then unrecorded issues (samples, damages, internal use), then quantities short-received at goods-in, and last of all unit-of-measure confusion. Unit errors are the sneakiest, because when cartons and pieces are mixed up the variance always lands on the same multiple. Rocketly holds stock in a single base unit while allowing alternative units per variant through Unit of Measure Conversion, so the confusion surfaces at entry rather than on the count sheet.

The work does not end at the adjustment. If the same SKU produces a variance in three consecutive sessions, the problem is not in the counting but in that item's flow. This is where threshold alerts that warn you before stock runs out earn their place: raising the threshold on a chronically variant line does not fix the cause, but it does reduce the stockout risk while you look for it.

How does a count variance land in the books?

A count variance is a stock movement, and like every stock movement it carries a cost effect. A shortage is multiplied by the item's current unit cost and lands as an expense; an overage increases inventory value by the same logic. So the cost basis you post the adjustment at matters as much as the count itself. The choice between FIFO and weighted average cost produces a different amount for the very same variance.

The bigger practical problem is the stock ledger and the accounting ledger drifting apart quietly. An invoice has been issued but stock never came down; or goods left the building and the invoice was raised later. Rocketly's Stock–Accounting Reconciliation screen lists records that were invoiced without a stock movement. Clearing that list before you open a session is what makes the variances you post genuinely physical. Otherwise you treat an accounting gap as a warehouse shortage, adjust it, and let two errors hide each other.

Which single number measures counting accuracy?

One indicator is enough: line accuracy. The number of lines whose variance stayed under the threshold, divided by the number of lines counted. Measuring by total units is misleading, because a single large discrepancy either disappears among hundreds of correct lines or single-handedly wrecks the picture. Measured by line, you get an answer to the question that matters: on how many lines out of a hundred can I trust the system?

The second indicator is time to resolution — days from detecting a variance to the adjustment being approved. If that stretches, counting is happening but producing nothing. The third is the number of repeat SKUs: lines that vary session after session. Tracking these three by warehouse, shift and class shows you which flow is broken rather than who is careless. That is the most concrete use of benchmarking against your own data: instead of hunting for an industry average, you compare two of your own warehouses on the same measure.

Rocketly's Stock Health Dashboard makes that tracking cheap: available stock, reserved quantity, consumption and dormancy signals sit in one table, and available stock that has gone negative is flagged separately. A negative line is a line that needs counting today, and you can start a count session directly from it.

How do you build the routine in ninety days?

Month one is preparation: every product gets an SKU and a barcode, warehouses and bin addresses are defined, opening quantities are posted as stock receipts. Set no accuracy target this month; the target is recording movements without delay. Month two is the pilot: A class only, one warehouse, one session a week, with the roles of generating, counting and approving separated from day one. Month three is rollout — B class enters the calendar, the second warehouse joins, and line accuracy is reported weekly.

Completing a large sheet can take minutes. Rocketly runs bulk updates, exports and large count completions through Background Job Tracking: queued, running, completed or failed status is visible with a progress percentage, and the job continues even if you close the browser. Nobody has to sit in front of a screen on count day.

Finally, do not treat counting as a warehouse-only concern. When orders, production and sales all read the same data, a count result converts immediately into promise capacity; a setup that brings stock, orders and production onto the same screen as sales ends the habit of the sales team phoning the warehouse to ask whether something is in.

If you want to try a first session in your own warehouse, you can open a Rocketly account and start with one warehouse and your A class items.

FAQ

Is shipping during a count really safe?

Yes, on one condition: shipments must be recorded as they happen. Because the session stores the baseline from the moment the sheet opened, later movements are shown separately and applied as an adjustment on completion. If recording lags, it is not safe — and the problem is the lag, not the count.

Do I still have to do a full annual count?

Statutory and audit requirements differ by business and jurisdiction, so settle that with your accountant. Operationally, in a warehouse that runs regular cycle counts the year-end count becomes a verification exercise: the variance list is short and the close is faster.

How do I set the variance threshold?

Set it per class and start from your own history. Look at the variance distribution across your first three sessions, take the level that roughly three quarters of lines fall under as the opening threshold, and tighten it in steps as accuracy improves.

Is a session too heavy for a small warehouse?

The opposite — a small warehouse is the easiest place to build one. In a hundred-SKU operation, counting twenty lines a week takes twenty minutes and adds up to five full sweeps a year. The heavy option is the three-day annual count that stops the business.