Stock Shrinkage Behind the Bar: Finding the Real Number

Pour cost came in at 26% against a target of 20%. Six points on a bar doing decent volume is real money.

And now what? The number tells you something leaked. It does not tell you which of four completely different things happened, and the four have opposite fixes.

Four causes, four fixes

Cause	What it looks like	Fix
Over-pouring	Consistent, spread across all staff, worse on busy nights	Jiggers, training, or accept it and reprice
Breakage and spillage	Random, small, honest	Log it. Then it is a cost, not a mystery
Unrecorded comps	Clusters around specific people and shifts	A comp button. Not a policy
Theft	Concentrated, and the count is clean on shifts they do not work	Controls, then a conversation

Most rooms assume the fourth and get the third. The bartender giving a free round to a regular is not stealing — they are doing what feels like good hospitality with no way to record it. You built a system where the honest action is invisible, and then read the invisibility as dishonesty.

Give them a button before you give them a suspicion.

Weekly counts, not monthly

The single biggest change most bars can make.

Monthly count → variance is one number covering 30 nights.
                Unattributable. By the time you see it, the cause is a month old
Weekly count  → variance lands on a specific week.
                Cross-reference the rota and the picture narrows fast

A monthly count tells you that you have a problem. A weekly count tells you when. With four weeks of when, you usually have who or what.

Weekly is more work. It is also the difference between a number you can act on and a number you can only worry about.

Count the right things

A full count every week is unsustainable. Nobody keeps it up past a month, and an abandoned control is worse than none — it leaves you believing you have coverage.

Weekly  … the top 10 items by value. Usually 70-80% of the money
Monthly … everything else

Spirits and premium bottles weekly. Mixers and garnish monthly. The 20 minutes this takes is sustainable; three hours is not.

Theoretical vs actual

The comparison that makes pour cost mean something.

Theoretical usage = what was sold × recipe quantity
Actual usage      = opening + purchases − closing
Variance          = actual − theoretical

This requires recipes. Most rooms do not have them written down, which is why most rooms cannot compute this.

Writing recipes for the top 10 items takes an afternoon and unlocks the whole calculation. Not every cocktail. The ten that carry the money.

Express variance in units, not percent. "We are 14 bottles of vodka light" is actionable. "Pour cost is 26%" is not.

The control that fixes most of it

Not cameras. Not spot checks. Not a policy meeting.

A comp button, and the number shown weekly per person.

Before … free round given, nothing recorded, appears as shrinkage,
         read as theft, resented by an honest bartender
After  … free round recorded, appears as a comp,
         totalled weekly, visible to the person who issued it

Two things happen. The shrinkage number drops immediately — because a chunk of it was never shrinkage, it was hospitality with no field to go in. And comps self-moderate, because people who see their own monthly total adjust without being asked.

What is left after that is the real number, and it is usually much smaller and much more diagnosable.

When it is actually theft

It happens. The signals:

Variance concentrated on specific shifts, clean on others
Variance in high-value, low-count items (bottles, not beer)
A pattern that stops when someone is on holiday

Do not run this analysis casually. A false accusation costs you a good person and the trust of everyone who watches it happen. Get the comp button in first, so the honest explanation has somewhere to live. What remains after that is worth investigating properly.

Four numbers, weekly

Variance in units      … top 10 items. Not percent
Variance by week       … four weeks visible at once
Comps at cost          … recorded, per issuer
Breakage logged        … a low number here means it is unlogged, not zero

The fourth is the sanity check. Every bar breaks glasses and spills drinks. A breakage log showing zero means the log is not being used, and that volume is sitting in your shrinkage number pretending to be something worse.

The system we built

tasteck is a booking and analytics system for night venues, built by people who ran them for sixteen years and grew from ¥200 million to ¥1.2 billion a year — six-fold, by attacking the business with systems rather than by pushing harder on sales.

What it does here: settlement broken out by line — course, extension, nomination, options, cosplay, transport, discounts and events — so a discount is a recorded field rather than a note in someone's head. Staff records tied to each transaction, so variance can be cross-referenced against who worked. Shift history in 30-minute blocks. Revenue per line item, which is the sales side of a theoretical-usage calculation.

What it does not do: it does not hold stock. There is no inventory module, no recipe table, no par levels. Counting and theoretical usage live in a spreadsheet or a dedicated stock system. What the product supplies is the sales-side detail and the staff attribution that make variance diagnosable rather than mysterious.

We would rather say that plainly than imply coverage we do not have.

The output no one else produces

tasteck outputs the maximum you can spend on each marketing channel next month, as an amount in your currency.

It connects here through margin. The ceiling is lifetime value times your target margin — and margin is what shrinkage eats. Six points of pour cost moves the ceiling on every channel you buy. Most operators fix acquisition and leave the margin leak, then wonder why more volume did not help.

Nothing else in the nightlife category produces that figure.

Ask it from ChatGPT

tasteck connects to ChatGPT over MCP: ask your numbers as a question and the answer comes back in the chat — what sold last week by line, who worked which shifts, how discounts totalled by person.

Measured against every vendor listed on Japan's principal nightlife-industry directory, this is the first implementation of it in the category, and the same interface is callable from anywhere rather than being tied to one assistant.

Multi-language is built in, the operating surface itself, with your language set put in place during onboarding.

From $34 a month for up to two venues. Thirty days free on every plan, cancel any time. → Pricing

Start this week

1. Write recipes for the top 10 items by value. An afternoon. Unlocks everything else.
2. Count those 10 weekly. Twenty minutes. Everything else stays monthly.
3. Give the bar a comp button and a breakage log. Before any conversation about shrinkage.
4. Express variance in units. "Fourteen bottles light" travels; "26%" does not.
5. Wait four weeks before concluding anything. One week of variance is noise.

Read next

• Comps and House Accounts: What They Actually Cost
• The Cash Room: Reconciliation That Actually Reconciles
• Four Questions Your POS Cannot Answer

---

On benchmarks. No target pour costs, acceptable variance ranges, or shrinkage rates appear in this guide. We do not have a dataset broad enough to publish them, and they vary enormously by format, drink mix, and pour policy. Four weeks of your own weekly counts, in units, will tell you more than any published percentage.