How to run a warehouse stocktake.

Planning and cut off

Start with the date and the cut off, not the headcount. A stocktake needs a single point in time at which every transaction, every receipt and every pick, has either happened or has not. Pick a cut off, publish it to receiving, picking, returns and any third party carriers moving stock in or out, and hold it. A count with a fuzzy cut off is not a count, it is an estimate with a clipboard.

Work backwards from the cut off to build the plan: which zones are counted in which order, how many people are needed per zone, what equipment they need, and how long each zone should take. Set a realistic pace rather than a hopeful one. Three to ten people counting a medium warehouse by hand typically need two to three days to complete a full count, and that figure should shape your schedule, not the other way round.

Decide in advance what "done" means for a zone: every location counted once, discrepancies flagged, and a supervisor sign off before the team moves on. Without that checkpoint, zones get half counted under time pressure and nobody notices until reconciliation.

Freezing movements, or not

The cleanest stocktake happens with all movement frozen: no receiving, no picking, no put away, for the duration of the count. That gives you a true snapshot, but it also means stopping the warehouse, which is expensive in lost throughput and unpopular with everyone who is not running the count.

Most sites compromise. Inbound and outbound are paused for the zones actively being counted and reopened once a zone is signed off, so the warehouse keeps moving overall while each individual count stays clean. If a full freeze is not possible, the discipline that matters is recording every movement that happens during the count, by zone and by time, so reconciliation can account for it rather than treating it as a discrepancy. An uncontrolled movement during a live count is the single most common reason a stocktake result gets challenged afterwards.

Zoning and teams

Split the warehouse into zones small enough that one team can count one zone within a shift, and assign one team, one supervisor and one zone at a time. Boundaries should follow the racking and aisle structure, not an arbitrary headcount split, so nobody is counting half an aisle and leaving the rest for someone else to remember.

Rotate teams away from the zones they normally work in wherever you can. Someone who picks an aisle every day already believes they know what is in it, and that familiarity is exactly what produces confirmation bias during a count: seeing what you expect to see rather than what is actually on the shelf.

Count sheets and scanners

Paper count sheets still work, but they are slow and they defer every transcription error to reconciliation, where it is much harder to trace back to the shelf it came from. A hand scanner removes transcription error and timestamps every count, but a scanner still needs a person to walk to each location, aim it, and confirm the read, which is why hand scanning tops out at roughly 50 to 100 units per hour per person.

Whichever method you use, capture location, item, quantity and unit of measure on every line, and capture empty locations as explicitly as full ones. A location nobody records is a location nobody can later prove was checked.

Blind versus informed counts

In a blind count, the counter sees no system quantity and simply records what is physically there. In an informed count, the system quantity is shown and the counter confirms or corrects it. Blind counts are slower, because there is nothing to anchor against. But they are far more resistant to the single biggest source of bad stocktake data: a counter who sees "12" on the sheet, glances at the shelf, sees roughly twelve, and writes down 12 without a careful count. Auditors generally prefer blind counts for exactly this reason, and most compliance frameworks treat an informed count as weaker evidence.

Recounts and tolerance

Decide your tolerance before you count, not after you see a number you do not like. A common approach sets a tolerance band, for example within one unit or within a small percentage of expected quantity depending on item value, and anything outside that band triggers an automatic recount by a second person before it is accepted. Anything still outside tolerance after a recount goes to reconciliation as a genuine discrepancy rather than a suspected miscount.

Recounting everything is not a rigorous stocktake, it is a stocktake that has run out of trust in its own first pass. Set the tolerance, apply it consistently, and let the data tell you which zones need a wider look.

Reconciliation and adjustment approval

Reconciliation compares the count to the system quantity, line by line, and produces the list that matters: what is genuinely different. Every difference needs a reason where one is knowable, whether that is a movement missed during the freeze, a receiving error found in the same run, or a real discrepancy with no obvious cause.

Adjustments to the system of record should never post automatically from a raw count. Route them through an approval step, normally the inventory or finance controller, so that a large or high value adjustment gets a second look before it changes the books. This is also the point where genuinely unexplained loss gets flagged for write off rather than quietly absorbed into next quarter's numbers.

The post mortem

A stocktake that ends at the adjustment file has wasted half its value. Hold a short review with the zone supervisors while the count is still fresh: which zones ran over time and why, which discrepancy types clustered where, and whether the cause was receiving, put away, picking or something upstream. That is the difference between a stocktake that fixes a number once a year and one that reduces how often the number is wrong at all. The causes of stock discrepancies works through each of those causes in turn.

Where automation changes the procedure

Automation does not remove the need for planning, cut off discipline or a reconciliation and approval step. Those stay. What changes is the middle of the procedure. A count that used three to ten people over two to three days can be run by one person in eight to twelve hours. A camera unit picked up by a forklift the site already owns reads every pallet label and every location, including empty ones, at up to 3,000 units per hour, and reconciles against the WMS automatically instead of by hand.

The six jobs one pass answers and wall to wall stocktaking set out the automated version of the same count.

Zoning still matters, because aisles still get driven one at a time, but count sheets, transcription and most of the recount cycle disappear, because every read comes with the camera recording behind it as evidence, so a disputed line gets settled by looking at the footage rather than sending someone to recount the shelf. That changes the economics enough that a full count stops being an annual event and starts being something you can run as often as the business needs. How often should you cycle count works through frequency, and the business case for stocktaking automation works through the arithmetic.