Every venue's stocktake comes out short and everyone assumes theft. Four causes account for most of it, only one of them is dishonesty, and the order you investigate them in decides whether you find anything.

The stocktake comes out short. It came out short last quarter too. Everybody has a theory, most of the theories are about people, and nothing changes.

The variance is real. What it means is almost always misread, because the four causes look identical on a spreadsheet and only one of them is dishonesty.

The four causes

1  Measurement — the count itself is wrong
2  Process     — over-pouring, spillage, comps not rung
3  Recording   — sales exist that the system never saw
4  Theft       — product or cash leaving

⚠️ Most venues investigate 4 first. It is the most expensive, the most damaging to morale, and statistically the least likely.

Why the order matters

Investigating theft first means accusing people before eliminating arithmetic. If it turns out to be a measurement error — and it often does — the accusation cannot be taken back.

⭕ Eliminate 1, then 2, then 3. Only what survives all three is 4.

Cause one: the count itself

More stocktakes fail here than anywhere else, and it produces variances large enough to look like theft.

Open bottles

Counted as full   — variance looks like a gain
Counted as empty  — variance looks like a loss
Estimated by eye  — error of 20-30% per bottle, in both directions

⚠️ With twelve open bottles on a back bar, eyeball estimation alone can produce a variance bigger than anything a person could carry out.

⭕ Weigh them, or use a consistent tenths scale, and use the same method every time.

Stock that is not where you count

Cellar, back bar, fridges, function room, the box behind the door

⇒ If a location is missed in one count and included in the next, you get a large loss followed by a large gain. Two impossible numbers, one missed shelf.

Deliveries that landed mid-count

A delivery signed in during the count is either double-counted or not counted at all.

⭕ Count when nothing moves. Not during trade, not on delivery day.

⇒ When The Supplier Changes Something

The count that changes people

If a different person counts each time, you are measuring the counters as much as the stock.

Cause two: process

This is where most genuine loss actually lives, and none of it is dishonest.

Over-pouring

A 25ml measure poured free-hand runs 10-20% over on a busy night

⇒ Across a weekend that is a meaningful quantity, and every single pour was made by someone trying to give good service.

⭕ Measure it: pour ten free-hand into a jigger and look. It is a five-minute test and it usually ends the theft theory on its own.

Spillage and breakage

Real, constant, and almost never recorded — which is why it shows up as an unexplained variance instead of a known cost.

⭕ A spillage line that takes three seconds to write turns a mystery into a number.

Comps and staff drinks

Approved but not rung through
Rung through as something else
Not rung through at all

⇒ The Comps You Never Approved

The tab that was never opened

A regular says "put it on my tab", no tab is opened, drinks are poured. The stock leaves and the sale does not exist.

⇒ The Bar Tab That Walks

Cause three: recording

Sales that happened but were never captured. The stock is correctly gone; the record is what is missing.

Terminal down and sales taken manually
A round rung as one item instead of four
Wrong product selected under pressure
Refunds and voids not reconciled

⚠️ A venue with frequent terminal problems will always have a stock variance, and it has nothing to do with anybody's honesty.

⇒ When The Card Machine Is The Queue

The wrong-button problem

Two products at the same price, one button used for both. Stock says one is disappearing and the other is not selling. Both numbers are wrong and neither is a person.

⭕ Look for a product that never varies and one that always does. That pairing is a button problem, not a theft problem.

Cause four: theft

What is left after the first three are eliminated. Two things are worth saying plainly.

It looks different

Process loss  — spread across many products, roughly proportional to volume
Theft         — concentrated: specific products, specific shifts, specific times

⚠️ A pattern that is concentrated is worth looking at. A pattern that is spread is a process pattern.

⚠️ What to do about it is not an operations question

Investigating an individual, what you may monitor, and what action you may take are governed by employment law and vary by country. Take it to your solicitor before you act. This page covers only how to tell the four causes apart.

Doing a stocktake that means something

Same person, same order, same method

Consistency matters more than accuracy. A count that is consistently 2% off still shows you the change; a count that varies in method shows you nothing.

Count often enough to be useful

Quarterly  — the variance covers three months, no way to attribute it
Monthly    — usable
Weekly on the top ten lines — where the answers actually are

⭕ You do not need to count everything often. Count the fast-moving, high-value lines weekly and everything else monthly.

⇒ Par Levels And The Money Sitting In Your Store

Write the variance reason at the time

The same discipline that makes a cash-up fast makes a stocktake meaningful.

⇒ The Cash-Up That Takes An Hour

What good looks like

Test one: can you explain last month's variance?

Working     — two or three named reasons covering most of it
Not working — one number and a shrug

Test two: is the variance stable?

A variance that is consistently 3% is a process you can work on. A variance that swings from 1% to 9% is a measurement problem — you are not measuring the same thing twice.

⚠️ Fix the swinging before you interpret the level.

Test three: has anyone poured ten measures into a jigger?

⇒ In most venues, nobody has. It is the cheapest test in this article and it reframes the whole conversation.

What this is worth

Variance % × cost of goods × annual volume

⇒ Even at a low percentage the annual figure is usually larger than owners expect, and the majority of it is recoverable through process rather than through investigation.

Counting the things that are not bottles

Most stocktake attention goes to spirits, and most avoidable loss is elsewhere.

Glassware

Breakage is constant, unrecorded, and replaced reactively

⚠️ A venue that runs out of a glass type mid-service has a stock problem that never appears in a stock report.

⇒ The Bar Back Nobody Scheduled

Draught

Line cleaning, wastage at the start of a keg, and the end of one are all real and all frequently uncounted.

⭕ If draught variance is consistently worse than bottles, that is a process answer, not a people answer.

Garnish, ice and consumables

Cheap per unit, large in aggregate, never counted

⇒ Par Levels And The Money Sitting In Your Store

What to do with the result

A variance that produces no action is a number somebody spent two hours generating for nothing.

Rank by value, not by percentage

A 20% variance on a slow-moving line — small money
A 3% variance on your best seller  — large money

⚠️ Percentage is what gets reported and value is what matters.

Pick one line and work it for a month

⭕ One product, one cause, one change. Then measure the same line again.

⛔ Attacking everything produces no attributable improvement and exhausts the goodwill for doing it at all.

Tell the bar what you found

Nobody told  — the count feels like surveillance
Told         — the count becomes a shared problem

⇒ The Report You Open Every Monday

The first stocktake after a change

Any change resets your baseline, and the first count afterwards is not comparable.

New menu ／ new supplier ／ new measure ／ new till buttons

⚠️ Expect the first count after a change to look wrong. It usually is not.

⭕ Note the change beside the count so next quarter's reader knows why the line moved.