Every Variance and No Score
Every count you run in our system records the variance on every line — the quantity out, the value out, held against a threshold that decides what needs review. Then the session closes and the numbers stop. Nothing turns a year of variances into an accuracy figure, a trend, or an answer to which location or which counter is drifting.
There is a difference between having the data and having the measurement, and it is easy to miss because the first one feels like the second. Our count sessions capture everything an accuracy programme needs. What they do not do is ever tell you how accurate you are.
The verdict
Counting stock is expensive, and the only justification for the expense is that it makes you more accurate over time. That requires a number, and a number requires somebody to compute it. We capture everything needed and compute none of it, which means our counting module supports the activity thoroughly and the programme not at all. If you count in one building you will not miss this. If you count across states with teams you rarely see, the accuracy figure is the entire point of the exercise, and you should ask us — and every vendor you are considering — to show you last quarter next to the one before it.
Run a count. Lines come back with a variance in units and in value. Anything above the threshold you set goes for review, and a variance that survives review becomes an adjustment somebody has to approve before stock moves. That is a good process and most of it is better than what it replaced.
Now run twelve counts across four locations over a year and ask a question that any operations manager will ask eventually: are we getting better? There is no screen that answers it, and no report you can run. The variances are all there, on individual sessions, one at a time, and nothing has ever added them up.
The arithmetic nobody is doing
Accuracy is not a difficult calculation. It is the count lines that matched, over the count lines you counted. What makes it valuable is not the sophistication — it is that the number is comparable across time, across sites and across people, which no individual session can ever be.
One quarter at one location — illustrative figures, not a customer's
Every figure above except the last two lines is already recorded by our count sessions today. The two that matter most are the ones nothing computes: the accuracy percentage, and the gross variance — because netting a shortfall against a surplus is how a stock problem disappears into a small number. A net variance of 48,200 on a gross of 271,900 is not a good quarter; it is two large errors cancelling, and the net figure is the one that will be reported.
A net variance near zero is the most reassuring number in stock control, and one of the least informative. Errors that cancel are still errors.
Why the missing measurement costs more at scale
A single warehouse with a few hundred lines does not need a scoreboard. The manager counts, sees the variances, and knows perfectly well which aisle and which shift the problems come from, because they were there.
Change two things — a catalogue in the tens of thousands of lines, and stock spread across states with separate teams — and personal knowledge stops scaling. Nobody has been in all the buildings. Nobody can hold the pattern. The only way to know whether the warehouse in one state is drifting while another is steady is to compute a comparable number for each, and that is precisely the thing the product does not do.
The variance data does not become less available at scale. It becomes less readable, which is the same thing as unavailable once nobody can hold it all in their head.
What a session tells you
- This line was out by seven units
- This variance is above the threshold and needs review
- This adjustment needs approval before stock moves
- This session cannot close until the lines are dealt with
- All true, all useful, all about today.
What a programme needs to know
- Are we more accurate than last quarter
- Which location is drifting and since when
- Which categories generate most of the error
- Are the same items wrong repeatedly, or different ones each time
- None of these is answerable from a session.
The right-hand column is what turns counting from a chore into a control. Without it, counting is an expense you repeat forever at a constant rate and never reduce, because reducing it requires knowing where the errors are concentrated — and concentration is a pattern across sessions, never inside one.
The one that is worth building even before accuracy
If we were only allowed one number it would not be the accuracy percentage. It would be repeat offenders: items whose count was wrong more than once in a period.
An item wrong once is noise — a miscount, a mispick, a unit-of-measure slip. The same item wrong in three consecutive counts is not noise; it is a structural problem with a cause you can find and remove, and it is usually one of a small number of things. A packaging change that made two products look alike. A location holding two similar items adjacent. A conversion applied in one direction only. Each of those is fixable once, permanently, and each is invisible in any single session.
What to demand of any stock system that claims a counting programme
These are in the order we would weight them, and the first two are the ones vendors are least ready for.
An accuracy figure, computed by the system, comparable across periods
Make them prove it: Ask for last quarter and the one before, on one screen. Not a session — a trend.
Gross variance reported alongside net
Make them prove it: Ask what happens when a shortfall and a surplus of equal size occur in the same count.
Repeat offenders — items wrong more than once
Make them prove it: Ask which items were wrong in more than one count this year.
Accuracy broken down by location
Make them prove it: Ask which site is worst and by how much.
Coverage — what has not been counted at all
Make them prove it: Ask which locations have gone uncounted this quarter.
Accuracy by counter or by shift
Make them prove it: Ask, then watch carefully. This one is genuinely double-edged.
The last one on that list, and why we would build it last
Accuracy by counter is the most requested version of this and the one we would be most careful with. It measures the person, not the process, and a counter who knows they are being scored has an incentive to report the expected quantity rather than the one in front of them — which destroys the only data you had. If it is built it should be visible to the person being measured, and it should sit beside location and category breakdowns rather than instead of them.
Three, and the data for all of them is already being captured
This is an unusually cheap set to build, because nothing new needs recording. Every figure below is computable from count lines that already exist.
Accuracy, with a trend and a location breakdown
Lines matched over lines counted, per period and per site, on one screen with history. This is the number the whole programme is supposed to be moving, and today nobody can see it.
Gross variance beside net
Reported together, always, because the gap between them is itself the finding. A quarter where they are close is a quarter with a few real errors; a quarter where net is small and gross is large is a quarter with a lot of them cancelling.
Repeat offenders
Items wrong in more than one count in a period, ranked. This is the one we would build first if forced to choose, because it converts a measurement into a specific list of things to go and fix.
None of these needs a new field. They are aggregations of data the counting module already writes on every session, which is why their absence is a reporting gap rather than a product gap — and why we would rather say that plainly than describe it as a roadmap item.
Talk to us about count accuracyWhat AWRA OpsHub does today
- Variance captured on every line, in both quantity and value, on every count session.
- A review threshold per plan, so material variances are separated from rounding and cannot be waved through.
- Variances raised as adjustments for approval rather than written straight to stock.
- A full audit trail per session — who counted, who reviewed, who approved, and every recount with its reason.
What it does not do
- No accuracy figure at all. Not per session, not per period, not per site. The percentage is never computed.
- No trend, so improvement or deterioration over time is invisible.
- No gross-versus-net variance reporting, so offsetting errors present as a small number.
- No repeat-offender view, which is the single most actionable thing the data could produce.
- No comparison across locations, categories, counters or shifts.
- No coverage report, so what has not been counted is not knowable in the product.
Not ours, by choice
- We publish no industry accuracy benchmark and would not trust one if we did. What matters is your own trend on your own data, and a vendor quoting a world-class figure at you is selling, not measuring.
- The worked example above uses illustrative numbers to show an arithmetic you can run on your own counts. It is not drawn from any customer.
Send us a year of count results
From whatever system you are on now. We will compute the accuracy trend, the gross-versus-net gap and the repeat offenders by hand and show you what the numbers say — which is a more useful first conversation than a demonstration of our screens.
Talk to us about counting