The Stock You Never Count

What falls outside the count

Glassware. Breakage is continuous and invisible until a busy night runs out of one type.

Linen and cloths. Consumed steadily, replaced in panic.

Small equipment. Batteries for the card terminals, printer rolls, bulbs, the specific cable.

Cleaning and back-of-house. Never in the stock system, always in the petty cash.

⚠️ None of these is individually expensive, and together they are a real line — bought at retail, at short notice, by whoever is free.

The three costs

The price. Emergency purchases cost more, always.

The time. Somebody's hour, usually a manager's, usually during service prep.

The night it ran out. ⚠️ The expensive one. A card terminal without a printer roll, a busy bar without the right glass — small failures with disproportionate effects at the wrong hour.

The two-tier count

Not everything needs weekly counting. Split it.

Tier one — counted weekly. Anything expensive, anything that stops service if it runs out. Short list, real discipline.

Tier two — counted monthly, roughly. Everything else. ⚠️ Approximate is fine — you are looking for "we are down to two boxes", not an exact figure.

A tier-two list that is counted roughly beats a comprehensive list that is never counted at all.

The par level

For each tier-two item, one number: the level at which you reorder.

Set it above what feels necessary, because the whole point is never reaching zero on a Saturday. The carrying cost of one extra box is trivial against one night without it.

⚠️ Without a par level, the count produces information and no action — someone sees there are three left and does nothing, because three does not feel urgent.

Breakage is data

Glass breakage in particular is worth recording, not to blame anyone but because the pattern is informative:

Concentrated at one station — a layout problem, usually fixable. Concentrated at close — a rushing problem, which is a staffing question. Concentrated on one type — the glass itself, and a different one solves it permanently.

⚠️ Recorded breakage is nearly always lower than actual breakage unless it is explicitly not a blame exercise. Say so, or the data will be wrong in a specific direction.

Who owns it

The single most common failure is that tier two belongs to nobody.

One person, one monthly slot, one list. ⚠️ Not "whoever notices" — that is the current system, and it is the thing being replaced.

Counting more often is not the answer

Weekly full count  accurate, and nobody sustains it
Monthly full count  sustained, and three weeks stale
Weekly on ten lines accurate where it matters, sustained

⚠️ Coverage costs sustainability. Pick the ten lines that carry the money and count those; let the rest run monthly.

Choosing the ten

Highest value × highest movement

⇒ ⚠️ Not the ones that go missing most. Those are usually cheap, which is why nobody guards them — and why they are not worth your count.

---

Variance is a signal, not a verdict

Consistent small variance  measurement, pouring, or spec
One large variance      an event; find it
Growing variance       a process changed, or a person did

⚠️ The first is the most common and the least investigated. If every count is off by roughly the same amount, your par or your spec is wrong — not your team.

The order to check

1. Is the spec what people actually pour?
2. Is the count including the same places every time?
3. Only then, anything else

⇒ ⚠️ Two catches more errors than three ever will. A shelf that is sometimes counted and sometimes not produces a variance that looks like loss.

---

What deliveries hide

Short delivery, signed for  becomes variance next count
Wrong item, accepted     becomes variance next count
Delivered to the wrong place becomes variance next count

⚠️ All three read as shrinkage a week later, by which point nobody can tell them apart from the real thing.

The thirty seconds that prevent it

Count the cases against the note before signing

⇒ ⚠️ Signing first and checking later is the same as not checking.

---

The conversation to avoid

⚠️ Do not present variance as an accusation. The moment counting feels like an investigation, the numbers start being managed rather than measured — and you lose the signal permanently.

Say  "the count is off; help me find where"
Not  "the count is off"

---

Stop counting things that do not move

Slow lines  count quarterly, or when you reorder

⚠️ Effort spent on the dusty shelf is effort not spent on the fast one, where the money actually leaks.

---

The bottles are counted every week. Nobody has ever counted the glasses.

They are replaced when someone notices there are not enough, which is always at the worst moment and always at the worst price.

---

Three to hold

Emergency purchases per month. Falls quickly once par levels exist, and it is the number that proves the change worked.

Items that ran out during service. Should be zero; usually is not, and usually nobody counts.

Breakage by type. ⚠️ Only useful if recording it is genuinely blameless.

Where the record has to sit

Tier two lives outside the stock system in most venues, which is precisely why it goes uncounted. It does not need a sophisticated system — it needs a list with an owner and a date, somewhere the person doing the count will actually look.

tasteck keeps recurring tasks on the staff-facing screens, so a monthly count is a task that appears rather than one somebody has to remember.

Nobody decided not to count the glasses. They just were never on any list.

---

Read next

• The Bottle You Bought For One Guest
• The Delivery Window Nobody Set
• The Weekend That Costs You Monday