Percent of what?

A friend of mine runs three cafés, and last spring she came out of a meeting worried. Someone had shown her a chart and said the words: coffee is nearly half of it, and it's falling fast.

Coffee falling by half sounds like the end of a café. She spent a fortnight on it: new beans, a new supplier, a price change.

The number was true. It was also coffee sold through delivery apps, which is 9% of what the shop takes. Coffee was 46% of that 9%, and in the business as a whole that was four pence in every pound. Meanwhile the drive-through, which is a quarter of everything, had been drifting down for months and nobody mentioned it.

Nobody lied, nobody made an arithmetic mistake, and the chart was correct. One question was never asked, and asking it takes ten seconds.

This piece is about that question, and the handful of habits that come with it. There is no software involved and no maths beyond addition.

Watch anyone dig into a business and you will see the same move over and over: take a total, break it into parts, look at the parts, pick one, and break that into parts.

Sales by how people order, delivery orders by what they buy, coffee sales by time of day. It is the same move at every level, and the whole of the analysis is a chain of them.

Which means the quality of the thinking rests on one thing: whether the parts still add up to the total.

That sounds too simple to matter. It is the check almost nobody runs, and it is the difference between a fact and a fiction that looks exactly like a fact.

Two of these make a split true and three make it useful. A split missing any of them will still draw a chart, and the chart will still convince a room.

Here is the café. Click into delivery apps, then use the toggle to change what the percentages are measured against. Same sales, same arithmetic, two completely different impressions of the shop.

Two things worth sitting with. The four parts of delivery add to 9.0%, which is exactly what delivery is worth. Go as deep as you like and the addition still works. That is the check from the last section, holding across a level.

Then press the toggle. Coffee is either 46% or 4.1%, and both are correct. One makes it the story, the other makes it a rounding error. My friend lost a fortnight to the gap between those two true sentences.

The five rules give you a map you can trust. They say nothing about where to walk, and a trustworthy map of somewhere irrelevant still wastes your morning.

Real businesses are lopsided. In the café, the counter and the drive-through are five sixths of everything, and the smallest slice is worth about a penny in the pound. So the decision about where to dig is made before you dig, and it is made almost entirely on size.

Two rules fall out of a picture like that.

• Never open a slice without knowing its size first. A dramatic collapse inside 1% of sales is a dramatic collapse inside 1% of sales.
• Something big moving a little usually beats something small moving a lot. The drive-through slipping two points is worth more money than office catering vanishing altogether.

The second one is hard, because small things moving violently are far more fun to look at. A slice that halves makes a better chart than a slice that softens by 3%, and the softening slice is often worth ten times as much.

That is not an argument for ignoring the small stuff. Delivery was worth looking at, and looking at it found something real. It is an argument for the size being on screen at the moment you choose, rather than worked out afterwards to justify a choice you already made.

Everything so far splits up a total as it stands today. The question people actually walk in with is a change: it went down, and why.

A change splits too, and the same rule applies: the pieces have to add up to the movement. But size and blame are unrelated. A big slice can explain nothing, because it did not move, and a tiny slice can explain everything, because it fell off a cliff.

There is a nastier version of this, worth knowing because it will one day be used on you. A total can move while every single part of it improves.

The café's average spend per order fell all year. Counter spend rose, delivery spend rose, and nobody got worse at anything. Delivery is cheaper per order and it grew, so the blend shifted, and the average fell on its own.

The defence is not cleverness. If the split covers everything and the shares are on screen next to the values, a shift in the blend is visible as a shift in the blend. Look only at the average and you cannot tell the difference between the business getting worse and the business changing shape. Those need opposite responses.

Nobody good asks one question. They ask four or five in a row, and each one is chosen by what the last one showed. Here is the café, the way it actually went.

The answer lives in question five, and you cannot get to question five from a standing start. You would never think to ask about spend per order until the first four had shown you that sales alone were hiding something.

Question three found nothing, and that was not waste. Ruling a slice out is how you earn the right to trust the slice you keep.

Real data is messy, and pretending otherwise is how methods like this get a bad name. Four things to watch for.

This works on a spreadsheet or anything else. The order matters: each step is only answerable once the one above it is settled.

Step seven is the gate. A conclusion that cannot say what it is a percentage of is not a conclusion. It is an impression, and impressions are what all this was meant to replace.

This matters more now than it did five years ago. A model will do any of this arithmetic faster and more accurately than you will. It will also do it happily on a broken split, and write the result up in clear, confident, well-organised prose.

There is a new trap in this, and it is the reason to be strict. Clear writing used to be weak evidence of clear thinking. Someone who could explain a breakdown well had usually looked at it. That link is gone. A confident, tidy answer to a badly framed question is now the cheapest thing in the building.

So the job of whoever reviews it moves up a level. Stop poking at the conclusion and poke at the split underneath. What is the total? What is this a percentage of? Do the parts add up? What is in the leftovers? Four questions, each answerable in seconds, and between them they catch more bad analysis than reading the summary twice ever will.

A split you can trust counts nothing twice, leaves nothing out, measures every share against the same total, uses an order you chose on purpose, and shows you time. Size tells you which part to open. Explaining a change is a separate job from describing a total, and averages need care.

None of that needs software. It needs the habit of not reading a percentage until you have asked what it is a percentage of. That takes about a week to become automatic.

What it buys is worth more than it sounds: the difference between having opinions about a business and knowing its shape. When anyone can produce a confident answer in four seconds, knowing the shape is the part that is still yours.