Counting Without Stopping Fulfilment
A dark store cannot close for a stock count, so it is counted in pieces while orders keep moving. The site is split into zones or slots; each one is locked to picking for a few minutes, counted blind, recounted where it disagrees with the system and unlocked again, and the whole store is covered over a rolling schedule rather than in one night. The method is a cycle count; what makes it work in a dark store is how the count is sequenced around live picking and how every movement during the count is captured.
The price of counting this way is the clean cut-off. Different slots are counted at different moments, so the result is never one snapshot of the store, and stock that moves while a slot is being counted can be counted twice or missed. A dark store's count plan exists to recover that: lock properly, record what moved, and reconcile at the slot boundary rather than assuming nothing moved.
Why a Full Count Does Not Fit
The promise
Delivery measured in minutesThe store sells speed. Pausing orders for a count is not a scheduling problem; it takes the store off the market for the duration.
The clock
No still periodMany sites trade for most of the day and take replenishment in the hours they do not trade. The quiet night a wall-to-wall count needs often does not exist.
The cost
Lost orders and lost customersA closed site loses the orders it would have filled and some of the customers who try a competitor instead. That second loss lasts after the count ends.
Against that, a rolling count is cheap, even if it is less tidy. It also suits the way dark-store stock goes wrong: errors build up slot by slot, every day, so counting a little every day finds them closer to when they happened.
Building the Count Plan: Zones, Slots and Velocity
A dark-store count plan is built from the order data, not the floor plan. Three decisions shape it.
- Choose the counting unit. A zone (an aisle or a temperature area) is simpler to lock and suits small sites. A slot (one storage location) disturbs picking least and suits sites where closing an aisle would block too many orders. Most stores use slots for fast lines and zones for slow ones.
- Set frequency by risk, not equally. Lines that are picked constantly, small and valuable lines, and short-life lines in the chiller or freezer gather errors fastest and are counted most often. Slow, bulky, low-value lines can wait much longer. The exact frequencies come from the site's own error history, and they are revisited when that history changes.
- Place counts in the quiet parts of the demand curve. Fast lines are counted in the lowest-order hour of the day; slow lines can be counted during peaks because almost no order needs them. The plan should also avoid the minutes after a replenishment lands, when slots are being filled.
Coverage is the test of the plan: over each full cycle every slot has to be reached at least once. A plan that keeps counting the same fast lines and never reaches the back aisle produces good numbers about half the store.
Running a Slot Count While Picking Continues
- Lock the slot in the system. Picker routing then avoids it. An order line that needs the locked slot is held for a few minutes, picked from an overflow location, or substituted, according to a rule set in advance, not by the counter on the spot.
- Freeze the system quantity at the moment of the lock. The comparison is with the quantity the system held when the slot was locked, not when the count is entered.
- Count blind. The counter scans or keys what is physically there without seeing the expected quantity. A counter who can see the expected figure tends to find it.
- Check expiry and condition while there. In chilled and short-life sections, note units at or past their sell-by date and damaged packs. They are counted, but separately from saleable stock.
- Recount every difference before unlocking. A second person recounts any slot that disagrees beyond the agreed tolerance. Only a difference that survives the recount is recorded as a variance.
- Unlock, then investigate before posting. Unlock the slot so picking resumes. The variance is posted only after someone has looked for the cause: a mis-pick into a neighbouring slot, a receipt booked to the wrong location, a unit sitting in returns.
Handling Movement During the Count
Movement is what breaks a rolling count. It cannot be stopped, so it is recorded and reconciled. Four situations cover most of it.
- A pick from a slot after it was counted, before the cycle closed. The count and the system now differ for a good reason. With the pick's timestamp, it is a reconciling item, not a variance.
Reconcile by timestamp - Replenishment arriving at a slot being counted. The new stock can be counted, missed or counted twice depending on timing. Hold put-away to locked slots; where that is impossible, log the receipt separately and add it after the count.
Hold put-away - Customer returns and rider returns. Units coming back from failed or rejected deliveries sit in a returns area until checked. They belong to no slot yet and are counted as their own location.
Count the returns area - Damaged or expired stock moved out. A unit pulled for write-off during the count leaves the slot but not the store. Count the damage bin too, or the store will look short.
Count the damage bin
A rolling count that treats every difference as a variance, without adjusting for known movement, produces noise. Within weeks the store stops believing it, and the programme continues as an activity rather than a control.
What a Rolling Count Proves, and What It Does Not
A rolling count produces different evidence from a single-date count, and reporting it as the same overstates it.
Reading the results of a dark-store cycle count
Each point follows from the method above.
What it shows
- Coverage over the cycle: the share of slots and of stock value reached, and when each slot was last counted
- Accuracy by slot and by product group, which is where the errors gather
- Location errors (right stock, wrong slot) separately from quantity errors
- Variances that survived a blind recount and an investigation
What it does not show
- The whole store's stock on one date: no single day was counted end to end
- A site-wide accuracy figure that means much: a blended number hides where the problem is
- Assurance, if slots were not truly locked or movement was not captured
- Independence, if the store's own pickers counted between orders
Which is why the counting team matters. A picker asked to count between orders will put the order first, and the count sheet will show it. Either dedicated counters do the work, or it is scheduled into the quiet window the site really has. How rolling counts compare with counting everything at a period end is covered in the guide to perpetual and periodic inventory systems, and how many times a year a multi-site chain is usually counted in full in the retail stock audit frequency guide.
An internal cycle count is still checked from outside when a lender holds a charge over the stock, when a network shrinkage figure is reported to investors or a parent company, or when the cycle has never been compared with an independent count. A stock audit for dark stores is planned around the store's operating hours so that the independent count does not stop orders either.
