Everybody quotes it and almost nobody can say what the denominator is. Divided by covers, by transactions, or by people who came through the door, the same night produces three different numbers and three different conclusions.

"Average spend is about forty" is the most quoted number in hospitality and one of the least examined. Ask what it is divided by and the answer is usually a pause.

That pause matters, because the same Saturday produces three quite different figures depending on the denominator — and the three point at different problems.

The three denominators

Same night, same takings. Different questions.

Divided by transactions

⇒ ⭕ What the till gives you without being asked. It measures the size of a purchase, not the value of a person, and it falls whenever people buy in smaller, more frequent rounds.

Divided by covers

⭕ People who were served. This is the one most operators mean when they say it, and it requires somebody to count heads rather than sales.

Divided by people who came in

⚠️ ⭕ Including everybody who bought nothing. This is the only version that measures the venue rather than the transaction, and almost nowhere calculates it.

⇒ The Door Count That Tells You Nothing

The same night, three numbers

Takings 8,400
　÷ 280 transactions … 30
　÷ 190 covers       … 44
　÷ 240 through the door … 35

⭕ Three answers to "what did a guest spend", all correct, none interchangeable. A venue comparing this month to last month using a different one has learned nothing.

Why it moves for reasons that are not spending

Four mechanisms that shift the number without anybody buying differently.

The round size changed

⇒ ⭕ One person buying for four looks like a large transaction; four people buying their own looks like four small ones. Identical revenue, very different transaction average.

The card machine changed behaviour

⚠️ ⭕ Contactless makes small separate purchases easy, which lowers the transaction average while raising the number of transactions. Nothing about the guests changed.

The mix moved between areas

⇒ The VIP Room That Earns Less Than the Floor

And the count of people is a guess

⚠️ ⭕ Covers are usually estimated and door counts are usually approximate. A denominator with fifteen per cent of slack produces an average with fifteen per cent of slack, which is larger than most of the changes people are trying to detect.

What it hides even when it is right

The bigger problem is that an average compresses the thing you actually need to see.

The distribution matters more than the middle

⭕ Forty as an average of forties and forty as an average of tens and hundreds are different venues. The second has a segment worth serving differently and the first does not.

⇒ What a Regular Is Actually Worth

The zero spenders are usually excluded

⚠️ ⭕ People who came in and bought nothing are silently dropped from most calculations. They are also the group most affected by anything you change about service or price.

It moves with who is in, not what you did

⇒ ⭕ A promoter night, a birthday, a group booking. The average follows the mix of the crowd far more strongly than it follows any decision you made.

And it says nothing about time

⇒ ⭕ Forty over two hours and forty over five hours are opposite results. One is a venue converting well and turning tables; the other is a venue with a table occupied all night.

⇒ The Table That Never Turns

The numbers that are more useful

Four replacements, each answering a question the average cannot.

Spend by hour of the night

⭕ The single most useful cut, because it is actionable. It tells you which hour is underperforming, which is a staffing and offer decision rather than an abstraction.

The proportion who bought a second

⚠️ ⭕ Whether people buy again is a different question from how much they spent once, and it responds to things you control — speed of service, where the bar queue is, how loud the room is.

⇒ The Volume Nobody Decides

The median, next to the mean

⇒ ⭕ Where they differ a lot, a small group is carrying the number. That gap is more informative than either figure alone and takes no extra data to produce.

And spend per person over a year, not a night

⇒ The Same Guest, Three Times

Making it comparable at all

If you keep the metric — and it is worth keeping — three rules make it mean something.

Write down the denominator, next to the number

⭕ ⚠️ "44 per cover" and "30 per transaction" are both fine; "about forty" is not. The label is what makes next month's comparison valid.

Never change it silently

⇒ ⭕ If the definition changes, the series breaks. Mark the month it changed on whatever you look at, or the change will eventually be read as a business result.

⇒ The Report You Open Every Monday

Keep the raw counts, not just the average

⚠️ ⭕ Takings, transactions, covers and door count, all four, every night. With those you can produce any version later; with only the average you can produce nothing.

And compare like nights

⇒ ⭕ A Saturday with a promoter and a Saturday without are not the same night type. Most month-on-month comparisons are comparing different mixtures.

⇒ Four Questions Your POS Cannot Answer

What to do when it drops

The number falls, and the instinct is to raise prices. Four things to check first.

Did the count change, or the takings

⇒ ⭕ An average can fall because the top went down or because the bottom got bigger. Busier with the same revenue is a different situation from quieter with less.

⇒ The People You Turned Away Last Saturday

Did the mix change

⚠️ ⭕ A different crowd on the same night of the week. Check the areas, the promoters, the group bookings before concluding anything about pricing.

Is anything physically slower

⭕ A queue, a card machine, a short-staffed hour. People buy less when buying is harder, and this shows up as an average spend problem while being a service problem.

And is the gap actually going somewhere else

⇒ The Pour Nobody Measures

What good looks like

The denominator test

Ask what the average spend figure is divided by. ⇒ A specific answer means the number is usable. A pause means it is a feeling with a decimal point.

The distribution test

⚠️ Ask what the median is alongside the mean. If nobody has it, nobody knows whether a few guests are carrying the venue.

The raw-count test

Working     — takings, transactions, covers and door count are all kept nightly
Not working — only the average survives, so no other version can ever be produced

Three to hold

1  Three denominators, three numbers, three conclusions — always label which
2  It moves with round size, card behaviour and crowd mix, none of which you did
3  Keep the raw counts; from those any version can be rebuilt, from an average none

Where the record has to sit

Takings, transaction count, covers and people through the door — four numbers, every night, in the same place. Kept there, you can produce whichever average a question needs, see the median beside the mean, and cut it by hour when something moves. Kept as a single average, the most quoted figure in the business is one nobody can define, comparable to nothing, and responsive mostly to who happened to come in.

⇒ The Thing You Measure Because It Is Easy

The Average Spend That Is Not An Average