What counts as unknown variance?

The difference between theoretical usage — what the sales mix says should have been poured — and actual usage from the count, after known losses such as spillage logged, comps and breakages are taken out. Under 2% of ex-tax cost of goods is the bar I hold myself to.

Is variance always theft?

Rarely, in my experience. Free-pouring, over-diluted or short batches, wrong specs learned from the wrong person, and till buttons that do not map to the product poured account for most of it. Fix those first; what remains is small and specific.

How often should a bar count stock?

Monthly for the full count, with high-value or fast-moving lines checked weekly. More often than that and the count becomes the job; less often and a problem has a quarter to grow.

Stocktake variance under 2% — what it actually takes

Jesse Williams · 18 September 2026

In one paragraph

Unknown stocktake variance — the gap between what the sales say you poured and what the count says you used — stays under 2% when four things hold at once — pars and ordering that keep stock levels boring, a prep programme where every batch is made to a written yield, specs the team actually pours to, and a monthly reconciliation that catches the drift before it compounds. I have held it at or under 2% at every venue since 2018, often within 1%, and none of it came from watching the team harder. It came from removing the reasons for the gap.

The count says you used it; the till says you didn’t sell it. The difference is variance, and every operator has a number they would rather not say out loud. Mine has been at or under 2% at every venue since 2018 — often within 1% — and I want to be precise about why, because the answer is not vigilance.

Variance is a symptom

A high variance figure is the bar telling you that one of its systems is broken. Which one is the whole question. In rough order of how often I find them:

The recipe is not the spec. A drink is built one way on the card and another way on the bar, because a bartender learned it from someone who learned it wrong. The till thinks 45 mL went out; 60 mL did.
The batch is not the recipe. A house cordial made to taste rather than to a written yield; a bottled cocktail diluted by eye. Every serve from that batch is mis-costed and the count cannot reconcile it.
The button is not the product. A till PLU never mapped to what it pours, so the sale is recorded against the wrong thing or nothing. Sales say one bottle moved; three did.
Ordering is chasing. Stock arrives in a rush because pars were never set, gets stored where it should not, and some of it walks.
Then, last, people. It is real, and it is usually smaller than the four above.

Watching harder addresses only the last. Fixing the first four addresses most of the number.

Pars and ordering: make stock boring

A par is the level a line is held at; a min/max is the band that triggers an order and caps it. Set from real weekly usage, not from the rep’s case deal, they do two things: the venue never runs out, and the venue never holds cash on the shelf it does not need. At the resort I run, pars and min/max thresholds across the high-volume outlets give full availability without tying cash up, and the order sheet is generated from them rather than from memory.

An order sheet that walks the shelves in the order they are laid out — with the supplier beside each line and a blank column to count into — is worth more than most software. The count becomes a walk, and the walk becomes the order.

The prep programme: every batch to a yield

Every prep and batch has a written recipe, a written yield and a written method. The prep sheet says what to make, how much it makes and what it should measure; the container label says what it is, when it was made and how strong. When a batch comes up short, the sheet has a column for it and the costing knows.

This is where clarified, fat-washed and infused preps earn their keep or lose it. A yield that lives in a bartender’s head drifts; a yield on a sheet that came from the costing database is the same on Tuesday as on Saturday.

Specs the team pours to

Spec cards that a new bartender can read on their phone in the walk-in, a blind-build test that checks they can pour it without the card, and product knowledge quizzes generated from the list the venue actually sells. Nothing in that is surveillance. It is the difference between one bar and eleven versions of it.

The monthly reconciliation

The count is set against the sales mix, theoretical usage against actual, per product. Lines outside tolerance get a name and a reason. Most months the reason is one of the four above and the fix is a specific small thing: remap a button, rewrite a yield, re-teach a spec. The number stays under 2% because the reasons for it to grow are removed one at a time, and because the whole exercise takes an afternoon rather than a weekend when the costing, the specs and the order sheet all come from one database.

What the number buys

A variance held under 2% is worth points of GP directly. It is also the number that lets an owner believe every other number — because a costing is only as credible as the count that confirms it.