Week 1: The Queue You Didn't Notice

Every queue you have ever stood in was designed. This week is about learning to see the decision instead of the line.

22 February 2027

Open the slides

There is a coffee shop on the ground floor of this building.

Between 9:00 and 9:20 on a teaching morning it serves about thirty people, on two members of staff, and for most of those twenty minutes the line is out of the door.

Ask the wrong question first

Most people, asked why that line exists, answer not enough staff.

Hold on to that answer. It is the one this entire course is built to complicate, and by week 4 you will be able to say precisely what is wrong with it. For now, notice only that it was reached in under a second, by everybody, without looking at anything.

Here is a better question, and it is not rhetorical:

Why is the line that shape?

Not why there is a line. Why it runs along the window rather than doubling back. Why there is one queue and not two. Why the person taking your order is also the person making the drink. Why somebody who wants a filter coffee waits behind somebody ordering four oat lattes.

Each of those is a decision. Not one of them was made by the people standing in the line, and most of them were not made by the people behind the counter either. They arrived with the fit-out, or with a franchise manual, or with a staff rota written to cover lunch.

Look at what the room is actually doing

Below is the same café three ways: same floor area, same two staff, same customers arriving at the same rate. The only thing the selector changes is how the room is organised. Run each one through the morning rush and watch where the line ends up.

ρ
—
L
—
W
—
W90
—
In line now
—
What to watch: Switch between the three arrangements at the same load. The first puts its tail out of the door, and the person at the counter is the same person making the drink. The second halves the line and lets one till stand idle beside a full one. The third has the same two staff as the second and a visibly shorter line, which is the part worth explaining. Along the bottom of the frame is the arrangement's rule, written out as an instruction — and the figure in pink is whoever that instruction has already chosen. Change the room and both change. In the two-line arrangement it chooses two people at once, and it will happily take the front of a line of one while a line of six waits.

None of these is the natural shape of a queue, because there is no natural shape of a queue. Each is a decision. Some of them were made deliberately by somebody who had thought about throughput. At least one was made by a shopfitter in 2019 who was thinking about where the power outlets go.

One of those three is costing about four minutes per customer at peak. Which, and why?

The first — one line, one person doing both jobs. At a rush of a customer and a half a minute it costs about 4.7 minutes each. The third arrangement, on the same two staff, costs 0.8.

Not because anybody is working harder. Taking an order is ten seconds and making the drink is twenty-five, so one person doing both spends thirty-five seconds a customer and the machine stands idle for ten of them. Split the two jobs across the same two people and the slow one never stops: the room’s capacity is now set by the drink alone.

Then push the rush past two a minute and something awkward happens. The third arrangement runs out of capacity before the second does, because two specialists in series can only go as fast as the slower of them, while two generalists share all the work. Two lines are worse at the rush the café actually gets and better at the one it does not.

Which is the real lesson, and it is not “this layout is best”: each of these rooms is optimal for a different morning, and nobody in that building could tell you which morning they chose, or when, or why.

Four parts, and only four

Strip any queue anywhere down and you find the same four parts. Almost everything in the next twelve weeks is a consequence of turning one of them.

Arrivals — when people turn up. In the café this is about ninety people an hour at peak, and the important thing about that number is not its size but its raggedness: a tutorial ends and eleven people arrive in forty seconds. Week 2 is about nothing else.

Service — how long each one takes. Ten seconds to take the order, twenty-five to make the drink, and a flat white is not a round of four. The spread matters more than the average, which is the other half of week 2.

Servers — how many at once. Two. It is the knob everyone reaches for first, it is the only one that costs money, and week 5 shows that how you arrange two servers changes the wait more than going from two to three does.

Discipline — who goes next. The café’s answer is “whoever has been in the line longest”, except when it is “whoever is holding the right card”, or “whoever the barista makes eye contact with”, or “whoever ordered something the machine is already making”.

The fourth one is the one nobody notices and everybody has opinions about. No one has ever been angry about inter-arrival variance. Everyone has been angry about queue-jumping.

Queues you would not call queues

Before we go further, widen the word, because the habit this course is trying to build is worth very little if it only fires in a coffee shop.

A queue is any place where work arrives faster than it can be done, and something has to decide the order. That is the whole definition. It does not require a floor, a rope, or a person standing on their feet.

Each of those has arrivals, service, servers and a discipline. Several of them have a discipline that has never been written down. The invisible ones are where the interesting failures live, because a rule nobody has stated is a rule nobody has argued with.

Where the mathematics came from, and who it was for

Worth knowing, because it explains the shape of the whole field.

In 1909, a Danish mathematician called Agner Krarup Erlang, working for the Copenhagen Telephone Company, published a paper on the theory of probabilities applied to telephone conversations. The company had a practical problem: how many lines does an exchange need so that callers are not turned away, given that callers arrive at random and talk for random lengths of time?

That is the founding question of queueing theory, and notice what it is not about. It is not about fairness, or whose call matters more, or who is standing in the rain. It is a capacity-planning question asked by an operator about a machine. The unit of offered traffic is still called the erlang, and the formulas he derived for how often a caller finds every line busy are still called Erlang B and Erlang C, and are still in use in every call centre in the world.

The mathematics of waiting was invented to size telephone exchanges, not to arbitrate between people.

Which is why the standard toolkit is superb at how long and nearly silent about who — it was built by operators, for operators, about equipment.

The orderly single-file line is younger than you would guess, and it is a social technology rather than a natural one. Crowds at markets and ticket windows for most of recorded history were a press, not a line. Queueing as a norm — the idea that the correct way to distribute a scarce thing is to stand behind the person who arrived before you, and that jumping is a moral offence — is generally traced to the industrial era, and in Britain it hardened into a national self-image during wartime rationing.

Which is to say: the rule you are about to spend four weeks taking apart was itself invented, quite recently, by people who had to be persuaded of it.

The claim this course makes

A queue is not a thing that happens to people. It is a rule, applied to people, by someone, usually invisibly.

Deciding who goes first is deciding whose time gets spent. Time is what everybody in that line is paying with, whether they agreed to or not.

So the rule is moving cost from one group of people to another — and it is almost never saying so, to people who have no way to bill anyone for the hours.

There is no neutral queue.

It will come back in week 4 with numbers attached, in week 8 with a price attached, and in week 12 attached to a transplant waiting list.

This is not a queueing theory course, though you will have the core of the mathematics by week 3 and it takes about forty minutes. The mathematics answers how long. It does not answer who.

The habit to build

Name the rule before you judge it.

A remarkable amount of complaint about waiting turns out, on inspection, to be complaint about a rule the complainer has never articulated. Once you can state it — as an instruction, beginning “when a server becomes free, serve…” — you can ask the question that actually goes somewhere: what would it cost to change?

The café above states its own. The sentence along the bottom of the frame is that arrangement’s rule written out as an instruction, and the figure in pink is the person the instruction has already picked; switch the room and watch both of them change.

Yours will be harder than that, because the queue you choose will not come with a selector. Three things go wrong, and all three are the point of this week’s lab:

You will write “first come, first served” and be wrong. Look again for thirty seconds and you will usually find a second rule running underneath: an app order that skips the line, a regular who gets served out of turn, a till that only takes cards, a member of staff who serves whoever is closest.

You will describe the intention, not the behaviour. The instruction has to be one a new starter could follow on their first shift without asking anybody what it means.

You will find nobody in the building knows. This is normal and it is the most interesting result you can get. A rule that has never been stated has never been defended, which means it has never been compared with the alternative.

How the next twelve weeks go

Every week runs the same shape, and it is deliberately not definition-then-explanation:

observation → model → experiment → consequence → design decision.

You will be shown a real place first, given the smallest model that explains it, handed the controls, and then asked what the result costs and who it lands on.

The semester is in three acts.

Act I — why waiting happens (weeks 1–4). The physics. You will learn to see a queue, explain why it forms when nobody is overloaded, measure one from outside with no cooperation from anybody, and put a price on the result.

Act II — who goes first (weeks 5–9). The rules. Five different ways to order the same people — arrival, length, need, price, appointment — and for each one, what it buys, and who pays for it.

Act III — systems that fight back (weeks 10–12). The complications. Queues joined to other queues, queues full of people who have worked out what you are doing, and finally a queue where the ordering rule had to be argued about in public because everyone could see what it decided.

Four pieces of assessment sit along that spine. You will measure a real queue (week 3), change a real rule and name the loser (week 6), run an experiment properly (week 9), and redesign something end to end (week 12). None of them asks you to recall a definition.

Where this goes

Next week, two systems. Both serve exactly sixty customers an hour. Both have one server working at exactly the same average speed.

One of them has no queue at all. The other has eleven people in it.

Nothing about how busy they are can explain the difference.

Teaching team