Seating by the window

Why does a nearly empty restaurant seat its first customers where everyone in the street can see them?

A restaurant opens for the evening with forty empty tables. Two people come in. Of all the places they could be put — the quiet corner, the banquette, the table by the kitchen nobody wants — they are put in the window, in full view of the street.

The standard explanation arrives immediately and sounds complete: an empty restaurant looks bad, a busy one looks good, so make it look busy. Probably true — but notice how weak the argument is. If a restaurant can manufacture the appearance of being busy at no cost, every restaurant can, and a signal everybody can fake should stop meaning anything to the people reading it. So either passers-by are naive in a way that never corrects, or the signal is not as cheap as it looks. Which is it?

Begin with the person on the pavement, because the whole thing is built around what they can and cannot know. They are hungry, they do not know this street, and they have a few seconds. What evidence about the food is available to them? Not the food, not the kitchen. The menu tells them prices and dishes but nothing about execution. Reviews might help, but consulting them costs attention and may not be trusted.

What is available instantly and for free is whether other people chose this place. That is not a foolish thing to attend to — other diners plausibly know something. So they use the only proxy they have, which is what makes it worth managing.

Now ask what the restaurant can do to that proxy. It cannot invent customers; it has whoever walked in. What it controls is which of its customers are visible. Seating from the window inward means the first two diners are seen by everyone passing; seating from the back means the same two are invisible and the room reads as empty. Same trade, opposite signal. The manipulation is not fabrication but a decision about the sampling window through which the street observes the room — and it can only shift the appearance forward in time, never past the number of people actually present.

That is the answer to the faking objection. The signal is not free and not unbounded. It requires real customers, it can be spent only once per customer, and it saturates: with two diners the choice of table changes everything the street sees, and with thirty it changes nothing. So the practice is concentrated exactly where you would predict — early in the evening, at the start of a season, in a new place — and disappears once the room fills. A cheap-signal objection predicts the practice everywhere or nowhere; the reality has a shape, and the shape follows the constraint.

There is a second thing to notice, which decides whether we have a mechanism or a story. If the point is signalling to the street, the practice should track visibility. A basement dining room, a restaurant behind frosted glass, a unit inside an airport concourse — none can send the signal, so if they still seat window-first we are wrong about why. Conversely a place trading almost entirely on bookings has little walk-in trade to influence, and some deliberately do the opposite, tucking early guests away because privacy is what their customers pay for. So the prediction is sharp: window-seeding should be strongest where walk-in trade dominates and street visibility is good, and absent where either fails. That is a claim you can walk down a street and check.

And here is where the tidy version deserves resisting. There are boring operational reasons to fill the front first that would produce the same behaviour with no signalling at all. An early shift often has one section open and one or two staff on; seating everyone in one zone shortens every walk, keeps the kitchen's tickets bunched rather than trickling, and lets the rest of the room stay unlit and unlaid. Customers also ask for window tables. So a fair share of what looks like clever psychology is scheduling — and I know of no clean field experiment measuring how much walk-in trade window-seeding actually produces. The mechanism and the incentive are real; the magnitude is not established.

Two refinements connect the steps, and one misconception is worth dismantling.

The misconception is that this works because people are gullible. The passer-by is not irrational; they are using the only evidence available at a cost they can afford, and occupancy really does correlate with quality on average. What the restaurant exploits is not a failure of reasoning but a gap between what the observer can see and what they want to know. Fill the gap — a trusted rating, a friend's recommendation, a booking made yesterday — and the window's influence shrinks, because a better signal displaces a weaker one.

The first refinement is about direction. A window table does not only signal outward; it also affects the diners in it, who may feel exposed or may enjoy being seen, and some venues seat forward for that reason alone.

The second is a boundary. This is not the phenomenon of travellers converging on one valley while an identical one stays empty. There the crowd is genuine, and the puzzle is that each person's rational deference stops their own choice carrying information, so a large crowd encodes only a few real judgements. Here there barely is a crowd. The subject is the party being judged arranging the evidence about itself, and the question is not why observers copy but why a manipulable signal retains any value — the answer being that it is bounded by real occupancy and spends itself as the room fills.

• Why visible queues work as signals in some trades and repel customers in others.

Every term the collection defines is gathered in the glossary.