The Count That Cannot See the Answer
A stock count where the counter can see the expected figure is not a count — it is a confirmation. What changes when the number is hidden, who is allowed to look, and why the audit trail on looking matters more than the hiding.
Here is a count sheet with a column headed "system quantity" and a column headed "counted quantity". A person walks the aisle with it, and in the second column they write, with remarkable consistency, whatever is printed in the first.
This is not usually dishonesty. Counting is tedious, the shelf is untidy, some of the stock is behind other stock, and there is a number right there that is probably correct. The mind supplies the answer before the eyes have finished the work — and it does so most reliably on exactly the items where a discrepancy would matter most, because those are the fast-moving ones that are hardest to count.
The fix is old and simple: take the answer off the sheet. What is worth thinking about is everything that has to be true around it for the fix to actually hold.
What a visible expected figure does to a count
Showing the expected number does not merely fail to help. It systematically biases the result, in a direction that makes the count worse than useless — because a count that agrees with the system is taken as evidence that the system is right.
- It anchors. Someone who counts thirty-eight and sees thirty-nine on the sheet recounts. Someone who counts thirty-nine and sees thirty-nine does not. The verification effort is applied only to the disagreements, which is precisely backwards.
- It converts a measurement into a check. The question stops being "how many are there" and becomes "is this right", and those produce different behaviour in front of an untidy shelf.
- It rewards the wrong outcome. In most workplaces a variance means paperwork, questions and possibly blame. Agreement means going home. The incentive points one way.
- It makes a genuine loss invisible. The item most likely to be quietly walking out of the store is the one where a matching count is most valuable to whoever is taking it.
- It destroys the evidential value of agreement. If the counter could see the figure, a count that matches tells you nothing at all — and matching counts are the overwhelming majority of any count.
If the counter could see the expected number, a matching count is not evidence of anything. And matching counts are almost all of them.
What blind counting means here
A count session can be marked blind when it is created. While that session is live, the recorded stock figure for the items in it is masked for the people doing the counting — the value is not merely hidden in the interface, it is removed from what the screen receives.
The important part is not the count screen. Any system can hide a column on a count sheet. The part that determines whether blind counting is real is what happens when the counter opens a different screen.
Here the mask follows the item. While a blind session is live, the stock figure is withheld on the item list and the item record too, on both the web application and the mobile app, for the people counting. Walking round the blind count by opening the item page in another tab does not work, which is the single most common way blind counting turns out to have been decorative.
| Surface | While a blind session is live |
|---|---|
| The count screen | Expected quantity withheld from the counter |
| The item list | Stock figure withheld and flagged as restricted |
| The item record | Same — withheld rather than displayed and ignored |
| The mobile app item endpoints | Same masking as the web application |
The mask applies to the person assigned to that count line, to ordinary inventory users on lines nobody has been assigned, and to whoever created the session — so setting up a count does not exempt you from it. It lifts automatically once the session leaves its active states, so there is nothing to remember to switch off afterwards.
Somebody has to be able to look
A control that nobody can bypass is usually a control that gets removed, because real situations turn up that it did not anticipate. A supervisor investigating a large variance genuinely needs to see the expected figure before deciding whether to recount, and refusing that outright would mean turning blind counting off on the day something interesting happened.
So there is a reveal, and it is held by administrators and by the people who hold the adjustment permissions. And every single use of it writes an audit event recording who looked, at which session and which line, for which item, and from which screen.
That trade is the actual design. Hiding the number is the easy part; making looking at it leave a mark is the part that changes behaviour, because a supervisor who reveals a figure knows the reveal is on the record next to the variance they subsequently approved.
The audit trail is the control, not the mask
If you take one thing from this page: a blind count where a supervisor can quietly look is not much better than an open one. A blind count where looking is recorded, and the record sits beside the variance approval, is a genuinely different thing — because now the question "did you check the number before you signed this off" has an answer that does not depend on anyone remembering.
What AWRA OpsHub does today
- A count session can be marked blind at creation, and the mask applies for as long as the session is open, counting, submitted or awaiting approval.
- The stock figure is withheld rather than hidden — the value is removed from what the screen receives, not styled out of view.
- The mask follows the item onto the item list and item record, on web and on mobile, so opening another screen does not reveal it.
- It applies to the session creator too, so setting up the count is not a way around it.
- Every reveal is written to an audit event with the actor, the session, the line, the item and the screen it was revealed from.
- It lifts automatically when the session leaves its active states — nothing to remember to undo.
What it does not do
- Stock reports do not mask. The masking is honoured by the count screens and the item surfaces; a stock report is a separate path and will show the figure. If your counters have reporting access, that is the gap to close with permissions rather than with the blind setting.
- There is no enforced separation between counting and approving. The system will let the same person be the counter on one session and the approver on another; whether that is acceptable is a policy you set with permissions.
Not ours, by choice
- The reveal permission is deliberately the same one that approves adjustments. Someone who can sign off a variance can see the figure behind it, and we would rather that be an explicit, audited pairing than pretend a supervisor can adjudicate a discrepancy blindfolded.
- We will not offer a silent reveal. There is no configuration that lets an authorised user look without the record being written, because the record is the entire point of allowing the reveal at all.
Masking on report surfaces, and an enforced separation between the counter and the approver on the same session, are both scope rather than ceilings. The visibility service, the permission model and the count audit log all exist and work — either is a written specification and a price.
The report gap is the one to act on before your first blind count. Check what your counting staff can reach in reporting, because a blind count is only as blind as the least restricted screen those people can open.
Running one so the result means something
The mechanism does its job in about a minute. Most of what makes a blind count worth doing is decided before anyone picks up a device.
-
Check what your counters can see elsewhere
Before the first session, open the application as one of the people who will be counting and try to find the stock figure — reports, exports, dashboards, the mobile app. The blind count is as blind as the loosest screen they can reach.
-
Count something worth counting
Blind counting costs more effort per line, so spend it where a discrepancy would matter: high value, high turnover, or the categories where shrinkage has previously shown up. Counting the whole catalogue blind, once a year, is worse than counting the top two hundred lines blind, monthly.
-
Assign lines to people
Assignment is what makes the mask precise, and it is also what makes the result attributable. An unassigned session masks more broadly and tells you less about where a pattern sits.
-
Decide the recount rule in advance
What size of variance triggers a recount, and by whom? Deciding after you see the number is how a recount becomes a negotiation. Write it down first — a percentage or a value, and a different person doing the second count.
-
Treat the reveal log as part of the review
When variances come up for approval, look at what was revealed and by whom. Not because reveals are suspicious — most are entirely proper — but because a control that nobody ever inspects stops being a control fairly quickly.
-
Investigate agreement too, occasionally
A department whose blind counts always agree exactly is either very well run or not counting. Both are worth knowing about, and a spot recount by a different person settles it.
What the variance is actually telling you
A blind count produces more discrepancies than an open one. This is the point, and it is also the reason blind counting gets quietly abandoned in month three — the variances look like a problem the count created rather than a problem the count found.
Worth saying plainly: the stock was already wrong. The count did not consume anything. What changed is that you now know, and knowing is what allows the far more useful question — which of these is theft, which is a receiving error, which is a sale recorded against the wrong item, and which is a unit-of-measure confusion that has been quietly running for a year.
Reading a variance before blaming it
- Is the difference a whole case, a half case, or a dozen? Round numbers point at a unit-of-measure or packaging problem rather than at loss.
- Is it one item or a family of similar items? A family points at a catalogue or receiving problem; a single item points at that item.
- Does it appear at one location only? Then it is about that location. Everywhere? Then it is about a process.
- Is it consistent in direction? Persistent shortages suggest loss or over-issuing; persistent surpluses usually mean stock is being received or issued in the wrong unit.
- Did the same variance appear last count? A repeating difference of the same size is almost always a systematic error rather than a series of coincidences.
- Was the expected figure revealed on this line before the count was submitted, and by whom?
The counting mechanics of doing this without shutting the shop are in counting the shop without closing it, the loss-pattern reading in how to stop stock shrinkage, and the approval side of a variance in reorder point and safety stock.
Our take
Hiding the expected figure is the easy half and every system offers some version of it. The two questions that determine whether it is worth anything are: can the counter find the number on a different screen, and does looking at it leave a record. Ours masks the item surfaces on web and mobile, not just the count sheet, and writes an audit event for every reveal — while stock reports remain a separate path you should close with permissions. Before your first blind count, log in as a counter and go looking for the number. Whatever you find is your real answer.
See blind counting in practice
Sessions marked blind at creation, the stock figure withheld across item screens on web and mobile, assignment down to the line, and an audit event for every reveal.
Explore inventory countingFrequently asked questions
What is a blind count?
A stock count where the person counting cannot see the quantity the system expects. The point is that a count taken with the answer visible is not really a measurement — it is a confirmation, and a confirmation that agrees tells you nothing, because agreement is what a person produces when they can see what agreement looks like. Marking a session blind here withholds the recorded figure from the counters for as long as the session is live, and lifts it automatically when the session closes.
Can a counter just look the item up on another screen?
Not on the item screens. While a blind session is live the stock figure is withheld on the item list and item record as well as on the count sheet, on both web and mobile, so opening another tab does not reveal it. Stock reports are a different matter — they are a separate path and do not mask, so if your counting staff have reporting access that is the gap to close with permissions. Log in as one of them and go looking before your first session; whatever you can find is what they can find.
Who can see the expected figure?
Administrators and users holding the adjustment permissions — which is deliberately the same group that approves the resulting variance. The reasoning is that a supervisor adjudicating a large discrepancy genuinely needs to see the number before deciding whether to order a recount, and a control that makes the sensible action impossible tends to get switched off. What makes the trade acceptable is that every reveal is recorded.
What exactly gets recorded when someone reveals a figure?
An audit event naming who looked, which count session and which line, which item, and which screen they revealed it from. That record is the actual control here rather than the masking. A blind count where a supervisor can quietly look is barely different from an open one; a blind count where looking sits on the record next to the variance they later approved is a different instrument. There is no setting that permits a silent reveal.
Our blind counts produce far more variances than our old counts. Is something wrong?
Almost certainly not — this is the expected and intended outcome. The stock was already wrong; the count did not create the difference, it revealed one that your previous counts were agreeing with. The useful work starts here: whole-case differences point at units of measure or packaging, a family of similar items points at a catalogue or receiving problem, a single location points at that location, and a variance that repeats at the same size every count is a systematic error rather than a run of bad luck.
Does the person who counts also approve the variance?
The system does not prevent it. Separating the two is a policy you set through permissions rather than something enforced on a per-session basis, so if segregation of duties matters to you — and in a shrinkage investigation it should — configure it deliberately rather than assuming. It is worth deciding alongside the recount rule: what size of variance triggers a second count, and by which different person.
